heatwavesandcoldspellsand …members.ziggo.nl › ekamper › publ › adh-2010-02-ekamper.pdf ·...

HEATWAVES AND COLD SPELLS AND

THEIR EFFECT ON MORTALITY:

AN ANALYSIS OF MICRO-DATA FOR THE NETHERLANDS

IN THE NINETEENTH AND TWENTIETH CENTURIES

by Peter EKAMPER, Frans VAN POPPEL,

Coen VAN DUIN and Kees MANDEMAKERS1

ANNALES DE DÉMOGRAPHIE HISTORIQUE 2010 n° 2 p. 55 à 104

55

INTRODUCTIONFor many centuries, Hippocrates’

thoughts on the role that weather condi-tions play on human health were asource of inspiration for physicians.Following the ideas that Hippocrateshad formulated in his On Airs, Waters,And Places, physicians studied the waysin which the natural conditions of acountry affected the appearance andvirulence of diseases. When vital regis-tration systems were introduced inEurope in the early nineteenth centuryand large-scale mortality data becameavailable, empirical studies on the effectof periods of extreme heat or coldbecame possible for the first time.Doctors were now able to identify thenegative health effects of heat and cold.For the Netherlands we can documentthe interest in this topic not only withstatistical studies and official reports butalso with literary sources and autobio-graphies, personal documents, newspa-per articles and iconographic sources.In 1868 summer temperatures in the

Netherlands reached extremely highlevels. With the help of locally collecteddata doctors studied the unusuallystrong increase in infant mortality

(Gedeputeerde Staten Zeeland, 1869;Godefroi, 1869). A leading Dutchhygienist, Casper Pieter Pous Koolhaas(1831-1893), stated that in manyplaces, in 1868 “more often than inmany other summers, in the summer ofthis year an infant’s body has beencarried to the grave”. The main reasonfor this extreme summer mortality wasthe kind of food supplied to children.Artificial feeding of children becameeven more difficult than it already wasunder normal circumstances. Duringhot weather, foods such as milk andbread porridge underwent “a slightchange, and in this process start to decay[…]. To dilute the milk, or to prepareother foods or drinks, water is used, andif one considers how poor the water is insome places as a consequence of the heatand drought, then one has anotherreason for the harms that are caused inparticular during hot summers by artifi-cial feeding of infants”. “The poorer thewater used, and the sooner and broaderthe decay of the food, the higher too isthe risk of uncleanness of the teats of thebottles in which one hands the food tothe children.” Pous Koolhaas arguedthat in hot summers the mortality ofchildren of the poor increased more

PETER EKAMPER, FRANS VAN POPPEL, COEN VAN DUIN AND KEES MANDEMAKERS

56

strongly than among children of thewell-to-do. “It is among these families inwhose dwellings, sooner than in thehouses of the rich, the bad air manifestsitself; in which, be it out of ignorance,negligence or frugality, more often bador badly prepared food is given to thechildren, or what has been left from anearlier day [...]. Lack of discernment,unfamiliarity with the need for it alsoleads in these families to a less-than-required care for the complete purity ofbottles and other utensils in which thechild’s food is kept and which is soextremely important, particularly in hotweather. The milk that is bought bythose people for whom spending a fewmore cents is a question of high impor-tance will generally be poorer in qualitythan the one that can be supplied bythose who wish to buy good stuff, evenif that costs a bit more; and those whobuy the bad and often-old milk, veryfrequently will have more trouble intrying to preserve it from decay.” (PousKoolhaas, 1869).The heat wave of 1911 drew even

more attention than the one of 1868, asthe latter’s death toll was indeed consi-derable. “Enormous loss of humanlives”, in particular among infants, “asad phenomenon, unparalleled in thestatistics of recent years”, “a massacre”especially in the countryside and less soin the larger towns “with their gooddrinking water and controlled milkstations”—in these terms the Dutchpress described the effect of the heat inNovember 1911 (De statistiek van denloop der bevolking, 1911). Medicaldoctors studied its consequences indetail, especially for infants. A surveyhad been set up in The Hague whichfollowed for several years all childrenborn in that city in 1908. The

researchers had ample opportunity topay attention to the effect of tempera-ture on the sampled infants (Gezond-heidscommissie 's-Gravenhage, 1913,72-89). They stressed that the effect ofheat periods was not only due to the risein temperature as such but alsodepended on the duration of the heatperiod. The authors observed thatmortality had increased not only forchildren aged one month or older butalso in the first month of life. Itappeared that after the extreme morta-lity in the summer months the deathrisks for infants in November andDecember were lower than in normalyears. Detailed daily temperature andmorbidity and mortality data wereanalyzed to examine the short- andlong-term effects of temperature onillness and death. The authors were ableto show that the number of days elap-sing between the heat peak and themortality peak shortened when the heatperiod kept up and temperatures rosefurther. Heynsius van den Berg (1912)found out through weekly data onnumbers of infant deaths that the largercities had withstood the summer of1911 relatively well, a finding that heexplained by the hygienic measures thathad been taken there in the recent past.He observed no direct adverse effect ofheat on infants; it was only after a seriesof days with extremely high tempera-tures that the situation for infantsbecame unbearable. This was a conse-quence of the fact that outdoor as wellas indoor temperatures reached veryhigh levels. Data for Amsterdam alsoshows the effect of increases in indoortemperature (De Lange, 1913). InSeptember 1911, when the heat hadalready disappeared but houses were stilloverheated, high mortality peaks could

HEAT WAVES AND COLD SPELLS AND THEIR EFFECT ON MORTALITY, XIXth AND XXth CENTURIES

57

still be found. De Lange observed thatthe heat had less of an effect on childrenyounger than one month. She creditsthis to very young children having rela-tively high rates of heat loss and the factthat many of these children were stillbeing breastfed. Many deceased childrenhad lived in single-room dwellings, andquite often the cooking and laundry wasdone in the room where the infantstayed during the daytime. Anotherproblem was that in periods of extremeheat thirsty children received solid foodor sour milk, and many infants weredressed with an excess of clothes (DeLange, 1913).Contemporary observers also noticed

the mortality-increasing effect of extremewinter temperatures. A case in point wasthe winter of 1890-1891, during whichthe agrarian population in particular, andamong them the poor in the first place,suffered. An anonymous letter to theeditor of a local newspaper (Het Nieuws-blad) on January 10th, 1891 described thesituation in a region called the HoekscheWaard, an island slightly north of theprovince of Zeeland. “The barren winterwhich quite unexpectedly holds swaywith implacably harshness takes a heavytoll among the well-to-do, but howmuchmore does it take out of our destitutehuman beings. […] Go and visit ourpoor and abide a few moments at thebedside of numbed old people; look atthe chilly, shivering children, coveredwith rags, yearning for a nourishing mealor a warming fire. Look at those toddlers,sleeping under pieces of rugs, floor mats,shredded clothes, or other things that acaring parental hand has been able tofind as cover; see with your own eyes thatin our midst people can be found whoare less well-off than animals in thecowshed; people who do not have a bed,

sometimes no straw, to lay their numbedlimbs down to rest, who are lackingeverything except distressing poverty,deep misery.” (Perneel, 2000, 249).After almost a century, in this paper

we study again the effect on mortality ofthe extreme Dutch winters andsummers of the nineteenth and twen-tieth centuries. We focus on those topicswhich already attracted the attention ofmedical doctors at the time: Were someage groups more vulnerable than others?Were there social classes that had toendure the effects of heat and cold morethan other groups? Were there indeedshort-term and longer-lasting effects ofheat and cold? In contrast to the nine-teenth-century studies, we combinedetailed mortality and temperature datawith advanced time-series methods toshed new light on the effects that heatwaves and cold spells had on mortalityin the past. Questions such as these haveonly received limited attention by histo-rians. Medical doctors and public healthspecialists had lost their interest in thetemperature-mortality link after WWI2,for various reasons. One is that water-and food-borne and air-borne infectiousdiseases, the spread of which was moresensitive to respectively extremely highand extremely low temperatures, losttheir importance. Deaths among infantsand children, the group most vulnerableto high and low temperatures, becamerare occurrences. And, very important,economic and technological develop-ments—improved housing, betterworking conditions, reduced outdooremployment, better transport, to nameonly a few—reduced exposure to andthe negative consequences of extremetemperatures. This all changed when inAugust 2003 Western Europe expe-rienced an unprecedented hot summer


58

(probably the hottest in Europe since1500; see Luterbacher, Dietrich,Xoplaki, Grosjean, & Wanner, 2004),with deadly consequences for thepopulation. Detailed analyses of theexcess mortality related to that heatwave were published for many Euro-pean countries (for France, see e. g.Rey, Fouillet, Jougla, & Hémon,2007). Haines et al. (2006) argued that“climatologists now consider it verylikely that changes in climate havedoubled the risk of a heat wave such asthat experienced in 2003”. We willattempt to show here that the relationbetween extreme weather conditionsand mortality and the changes thereinover time are an interesting researchtopic for historians too. How societiescoped with environmental shocks suchas extremely high temperatures andwhether or not they were able torestrict their effects on mortalityprovides us with a valuable measure ofsocietal development (Galloway,1994). Studying the differences invulnerability to this environmentalstress by social class, age and sex givesus information on the conditions—food, shelter, clothing—under whichthese groups lived and on the way theseconditions changed over time (Bengts-son, 2004). This is nowhere betterillustrated than in Eric Klinenberg’sHeat wave: a social autopsy of disaster inChicago (Klinenberg, 2002).Compared with a recently published

study in which we also analyzed thechanging temperature-mortality link,the present study encompasses data forfour provinces instead of the single onethat was the topic of our earlier paper(Ekamper, Van Poppel, Van Duin, &Garssen, 2009). This not only allows usto reach firmer conclusions, it also

makes it possible to compare regionsthat differ in their microclimatologicalenvironment.

THE TEMPERATURE-MORTALITYLINK AS TOPIC FOR HISTORIANS

There have been few quantitativeassessments of excess mortality duringheat waves or cold spells for historicalpopulations, let alone studies in whichthe vulnerability of specific groups wasanalyzed. Historical studies of the linkbetween extreme weather conditionsand mortality are, as a rule, based onrather crude weather and mortalityindicators. Mortality is mostly availableonly for the population as a whole,without distinction by sex or age, andoften only on a monthly basis, whereastemperature data tend to be monthlyaverages (see e. g. Galloway, 1985;Galloway, 1986; Galloway, 1988; 1994;Landers, 1986; McDowall, 1981). Animportant drawback of a monthlyaggregation of mortality data is that itmakes it almost impossible to properlyidentify the effects of extreme heat orcold on mortality. Deaths caused byextreme heat, for example, appear tooccur at very short lags (0-1 days) and itis therefore highly likely that an effectof heat-related mortality will be atte-nuated even in a weekly-aggregatedanalysis (Carson, Hajat, Armstrong, &Wilkinson, 2006). Historians rarelyhave access to information on dailynumbers of deaths and daily tempera-ture data, and where that is the casethese records are collected only forsmall communities and restricted timeperiods.The number of studies in which the

relation between extreme weatherconditions and mortality is studied over


59

a long period of time with adequatemethods and on the basis of comparabledata is extremely limited. Contempo-rary epidemiological studies rarely covera period long enough to encompassmajor economic, demographic orepidemiological transitions (Carson etal., 2006). There is not a single study inwhich advanced time-series methodshave been used to study mortalitydisplacement during heat waves andcold spells for historical populations.Recently, Rau (2007, 13-14) noted that“surprisingly, there is not much litera-ture in the field of seasonal mortality onthe ‘classical’ social mortality determi-nants such as income, deprivation,wealth, marital status, education, occu-pation”. He also noted that “most ofthese analyses […] studied the samecountry (UK) using similar methodsbased on ecological data”. Rau was refer-ring to the present-day situation, but hisconclusion applies even more to thehistorical study of the effect of extremeweather conditions. Although duringheat waves and cold spells contempo-raries often referred to the effects thatextreme weather had on the poor inparticular, there is hardly an empiricalstudy focusing on this aspect.In past decades historical databases

with more detailed information ondeaths have become available for anumber of countries. This paper usesdata relating to four of the elevenprovinces of the Netherlands, covering aperiod of 100 years and allowing us tostudy the effect of extreme temperaturesseparately by age, sex and social class. Werelate these mortality data to series ofstandardized, location-specific dailytemperature measurements, and in doingso we make use of sophisticated statisticalmethods. The data cover the period

during which the Netherlands under-went a transition from a mortality regimecharacterized by high annual fluctuationsin mortality due to the dominance ofinfectious diseases (lasting until around1875) to a regime in which infectiousdiseases disappeared almost completelyand degenerative diseases became themost important cause of death. Thistransition of the cause-of-death patternwas accompanied by a changing ageprofile of death, in which no longerinfants but the highest age groupsaccounted for the majority of deaths.The period that we study is also an

interesting one because it was characte-rized by strong socioeconomic progress,which might have caused a reduction inthe vulnerability of the population toexternal circumstances: national incomegrew rapidly after 1860, housing condi-tions improved, clothing became better,and food and fuel became widely avai-lable. As we are able to compare regionswith different levels of economic deve-lopment we have an excellent opportu-nity to find out how vulnerability toextreme circumstances changed overtime and varied by region.Of course our dataset also has some

drawbacks. Periods of extreme heat andextreme cold are scarce in theNetherlands,with maximum temperatures onlysporadically exceeding 27°C and mini-mum temperatures rarely droppingbelow –10°C. As a consequence, the va-riation in temperature-related mortalityis small by international standards(Healy, 2003; Keatinge et al., 1997;Keatinge et al., 2000; McKee, 1989). Onthe other hand, studies have documentedthat in countries with harsh climaticconditions during winter, winter excessmortality is lower than in countries withrelatively warm or moderate climates,


60

and this same mechanism applies to theexcess mortality during summer. This“seasonality paradox” (Gemmell,McLoone, Boddy, Dickinson, & Watt,2000), resulting from the fact that thepopulation is not accustomed to protec-ting itself adequately from uncommontemperatures, might have led to strongeffects even in a country with a moderateclimate like the Netherlands.Unfortunately, we have no information

on climatic conditions other than tempe-rature which might have an effect onmortality, such as humidity, wind speedor wind direction. Nor do we have infor-mation on temperature-related variablessuch as air pollution and influenza, whichmight have played a role in (changes in)weather-related excess mortality.

MECHANISMS INWEATHER-RELATED MORTALITY

A large number of studies presentoverviews of the factors that account foran increase in mortality due to cold orheat in contemporary societies. AsKeatinge and Donaldson (2004, 1094-1095) have made clear, few of the excessdeaths during cold are due to the bodysimply cooling until vital organs such asthe heart cease to function, and few heat-related deaths are due to hyperthermia,overheating of the body. Cold-relateddeaths are mainly caused by coronary andcerebral thrombosis and respiratorydiseases, whereas the same thrombosesaccount for most heat-related deaths. Theprecise effects of extreme heat or colddepend not on temperature as such alone3but also on specific conditions in whichthe temperature decline or rise took placeand on other climatic conditions.The effects of cold and heat may

consist of a more or less instantaneous

effect and a more delayed effect. Tempera-ture falls in winter are closely followedby increased mortality, with characteris-tic time courses for different causes ofdeath. For heat periods too, immediateeffects (such as acute myocardial infarc-tion) as well as long lag times might bedistinguished. The length of the period ofheat and cold might be a factor deter-mining the effect on mortality. For heatand for cold it might be assumed thatthe effect on mortality is higher thelonger the period during which thetemperature is extreme (Huynen,Martens, Schram, Weijenberg, &Kunst, 2001). Main heat effects areusually visible on the current day or maylast another day or two (Pattenden,Nikiforov, & Armstrong, 2003).Compensatory effects on mortality mightbe registered when longer time periodsare studied. The number of deathscaused by heat waves is often assumed tobe compensated for by a fall in numberof deaths in subsequent weeks. Thesuggestion is that heat mainly has aneffect on people whose health is alreadyimpaired and who would have diedwithin a short time anyway. Thiscompensating effect is known as“harvesting” effect (Huynen et al.,2001). However, no general agreementexists among scientists on the length ofthe period over which harvesting effectscan be expected, which vary from a fewdays or weeks in the short term toseveral months or even years in thelonger run (Toulemon & Barbieri,2008). The effects of heat and coldmight also be contingent on the suddenoccurrence of a change in temperature.Such effects depend on whether popula-tions have had the chance to adapt toextreme weather conditions (Ballester,Michelozzi, & Iniguez, 2003). Effects of


61

outdoor air temperature might be modi-fied by other weather conditions, such ashigh humidity and strong air flow (Gill,Davies, Gill, & Beevers, 1988). A studyon daily variation in mortality in rela-tion to temperature and two wind-chillindices for the Netherlands (1979-1987) showed that hazardous weathersituations could be identified almost asaccurately by temperature as by an indexthat also included wind-chill (Kunst,Groenhof, & Mackenbach, 1994).Effects of high and low temperaturesalso depend on the climatological situa-tion of the region studied. Studies ofpopulations living in widely differentclimates show that they have adjusted totheir own climate remarkably effectivelyover time. This applies to cold as well asto hot regions (Keatinge et al., 2000).Countries with the mildest winterclimates exhibit the highest effect inwinter mortality (Healy, 2003). Breschiand Livi-Bacci (1994) and Oris et al.(2004, 392-393) showed that winterpeaks in mortality among infants weremore common in climates with mildwinters than in harsh climates where thepopulation had a high capacity for adap-tation: thus in temperate regions winteris a more dangerous and impactingperiod than summer, although clothes,heating and good housing could reduceits effects.It is important to stress that the

temperature-mortality link might be dueto mechanisms other than the direct effectsof exposure of the human body to extremetemperatures. In particular for historicalpopulations, these indirect effects onmortality cannot be neglected. Wemention here two of these mechanisms.Extreme weather has a direct effect onbiological processes that are crucial toman’s survival, such as the growth of

foodplants and animals, and on the phy-sical environment, such as flooding andstorms. These direct effects could lead tosecond-order effects on mortality(Michaelowa, 2001). Relevant is also thelink between temperature and the inci-dence and virulence of infectiousdiseases, the most important cause ofdeath until the first decades of the twen-tieth century. Temperature and rainfallaffect the mobility and strength of patho-genic micro-organisms and those ofinsects and animals that carry them.Where sanitation was virtually unknownand water supplies subject to contamina-tion, warm summers promoted thespread of infectious diseases throughincreased proliferation of animal, insectand bacterial vectors (Galloway, 1994).Cases in point are malaria, diarrhea andother gastric conditions, the latter parti-cularly affecting those children who hadlost the protection of the mother’s milk(Oris et al., 2004).In studies dealing with present-day

effects of extreme temperatures onmortality the question often is whetherthere are specific groups whose health ismore affected by extreme heat or coldthan others. Usually the focus lies ongender and/or specific age groups, andphysiological factors are used to explaindifferences. Significant variations ineffects of heat and cold according to agehave been related to variations in ther-moregulatory function and appreciationof cold and heat with age. This isconsidered the main reason why theelderly are disproportionally affected byextreme weather conditions (Hajat,Kovats, & Lachowycz, 2007). There isno evidence in present-day studies ofexcess mortality attributable to heatwaves in children (Kovats & Kristie,2006), and only rarely is mention made


62

of the effect of extreme cold on thedeath risks for this age group. The deter-mination of which gender is moresusceptible to weather fluctuations ismuch in dispute. In studies of England,Wales and France women had higherheat-related mortality, reflecting adverseeffects of menopause on thermoregula-tion (Hajat et al., 2007; Rey et al.,2007). For cold-related mortality,gender differences were not significant(Keatinge et al., 1997).Relatively little present-day research has

examined variation in temperature vulne-rability by socioeconomic position, and thefew existing studies often present conflic-ting results. O’Neill et al. (O'Neill,Zanobetti, & Schwartz, 2003) observedstronger cold and heat effects among theless-educated inmost of the sevenUS citiesthey studied. Such an effect was not foundin a Spanish study (Borrell et al., 2006).Naughton et al. (2002) found increasedrisk of heat-related death during the 1995Chicago heat wave among low-incomeresidents, whereas Kaiser et al. (2007)found the same effect among the lowereducated. McDowall (1981) observedhigher winter excess mortality in Englandduring the 1959-1972 period among semi-skilled and unskilled workers than amongother social classes. Donaldson andKeatinge (2003) observed for 1998-2000in England and Wales that cold-relatedmortality in men of working age was lowfor unskilled occupations but high amongmen of retired ages in that same socialclass. The beneficial effect of work-relatedfactors in this social class was explained byinternal heat production from manualwork, offering protection against daytimecold stress. Other authors have arguedthat unacceptable working conditionsduring high temperature periods can leadto increased mortality in lower social

classes. Rau (2007, 127-162) studiedindividual-level data for Denmark for1980-1998, using a variety of socioeco-nomic indicators. He did not observe aconnection between excess winter morta-lity among people aged 65 years or olderand factors such as educational level,wealth and housing conditions. Areastudies give conflicting results too. Astudy for the 1993-2003 period in theUK (Hajat et al., 2007) observed verylittle difference in heat effects accordingto level of deprivation of the neighbor-hood and no link between cold and depri-vation. Results of a study of the French2003 heat wave however point to themost deprived populations being morevulnerable to heat waves (Rey et al.,2009). It remains to be seen whether inthe nineteenth and early twentiethcenturies such a difference in vulnerabilityalso can be observed.

SETTING, DATA AND METHODS

Study regions

Nationwide and compulsory birth anddeath registration according to the ruleslaid out in the Napoleonic Code wasintroduced in the Netherlands in 1811, atthe time of incorporation of the Nether-lands into the French Empire. In recentdecades, dozens of staff and volunteers inDutch provincial archives have started toenter death records into a database withinthe framework of projects called ISIS andGENLIAS. The purpose of these projectsis to build a database with genealogicalinformation on all marriages, deaths andbirths taking place in the Netherlandsfrom the introduction of the vital registra-tion system (1811) up to when such datawere not yet in the public domain. Death

Fig. 1 Map of the Selected Dutch Provinces around 1920

The four provinces each have theirown particular ecological, social andeconomic structure. Gelderland islocated in the central eastern part of thecountry, extending from the Germanborder westward to the former ZuyderZee. In the northwest the hill plateau ofthe Veluwe was a wasteland coveredwith heath and some woods, ill-adaptedfor cultivation and of little economicvalue except for some wood-cutting andpaper mills. The fertile marshy area ofthe Betuwe between the Rhine and theWaal supported orchards, marketgardening and mixed farming. Thesouthwestern section was a long, narrowwestward extension along the Rhineriver with brickyards and dairy farming.

Some textile works were located to theeast. Small regional marketplaces andseveral larger towns such as Arnhem andNijmegen hosted industrial activitiesand administrative services. Farms inGelderland were relatively small, theinfrastructure less well-developed, andthe productivity of land and labor lowerthan that of the coastal provinces.Drenthe is located in the northern part

of the country and shares an easternborder with Germany. The soil consistsalmost entirely of sand and gravel, andwas for a long time covered with bleakmoorland, patches of wood, and fen.Cultivation of buckwheat and peat-digging took place on the barren heathsand sodden fens found on the sand


63

records enter the public domain after 50years. We were able to use data for four ofthe eleven provinces of the Netherlands:Zeeland, Drenthe, Groningen andGelderland.These provinces were selectedbecause entry of death certificates has

been completed for the whole of theprovince and because the informationentered in the database includes informa-tion on sex, age and occupation of thedeceased4. Figure 1 gives an overview ofthe location of the selected provinces.

Drenthe Gelderland Groningen ZeelandPeriod M F M F M F M F1850-59 41.2 41.4 41.9 43.4 39.1 41.1 30.1 31.81901-02 49.5 50.2 50.2 52.4 50.4 52.9 52.0 55.41956-60 72.3 75.0 71.8 74.6 71.8 75.5 72.6 75.2

Tab. 1 Expectation of Life at Birth, by Sex, Period and Province

Source: Calculations by the authors derived from analyses of data on age and sex structure by provinces at census dates and numbers ofdeaths by age and sex from vital registration.


64

grounds, where sheep and cattle werereared and forest cultivated. In connec-tion with the cultivation of potatoes,factories were founded for making spirits,straw paper, etc. The people of Drenthehave had a long history of poverty. Thepoor agricultural soil did not always yieldenough to prevent farmers from starving.People often lived in turf huts andsupplemented their incomes throughpeat-cutting. Besides that, a large penalestablishment was erected in the mid-nineteenth century to which drunkardsand beggars from all over the countrywere sent. Owing to its geographicalisolation, the development of theprovince remained behind that of allother Dutch provinces. There were fewurban centers of any importance andpopulation density was rather low.Zeeland forms the southwestern part

of the coastal zone and consists of a stripof the Flanders mainland, borderingBelgium and six former islands, all ofthem now connected to each other or tothe inland provinces by dams andbridges. Much of Zeeland was below sealevel and protected by a system of riverand sea dikes. For a long time, Zeelandwas a rural area with the towns ofMiddelburg and Vlissingen as adminis-trative and industrial centers. Grainfarming on the sea clay was the chiefeconomic activity (60 per cent of thelabor force was involved in agriculture).

Part of the population was active in thefishing industry. In the second half ofthe nineteenth century agriculturalmodernization was eroding the positionof farm laborers. The economy of theregion started to change after 1900 withindustrialization (Priester, 1998;Wintle,1985).Groningen, situated in the extreme

north-east of the Netherlands, can beroughly divided into two regions: anorthern area of clay soils and a sou-thern one of sand and peat. The peatdistricts became an area of importantindustrial development in the secondhalf of the nineteenth century. Acommon feature of the agriculture ofboth areas was the high degree ofcommercialization.Mortality levels in the four provinces

differed considerably. Whereas Zeeland,like other Dutch coastal and low-lyingareas, was characterized by very highmortality until late in the nineteenthcentury—particularly among infants,reaching levels of 350 deaths before age1 per thousand live births—Drenthe,Groningen and Gelderland were doingmuch better. As Table 1 shows, theexpectation of life at birth in Zeelandwas much lower than elsewhere until themiddle of the nineteenth century, and itwas only in the century’s latter decadesthat the province reached higher valuesof life expectancy.


65

Together, the four regions cover a largeportion of the economic, demographicand cultural landscape of nineteenth-century Netherlands. The climatologicalconditions in the four regions differedslightly. The average annual temperaturein Zeeland was usually slightly higherthan elsewhere in the country, due to itscoastal location and a higher number ofhours of sunshine. In particular, theaverage minimum temperature washigher in Zeeland than in other parts ofthe Netherlands, with temperaturesrarely dropping below –10°C, especiallyin the westernmost regions. In Gelderlandhowever, and even more so in Drentheand Groningen, minimum and maxi-mum temperatures were more extreme.The average number of frost days (belowfreezing point) in the northeasternprovinces was much higher than inGelderland and Zeeland (Heijboer &Nellestijn, 2002).

Weather and mortality data

Temperature readings were taken atdifferent weather stations in the Nether-lands starting in the early 1850s. At thebeginning of 2000, the Royal Nether-lands Meteorological Institute (KNMI)started research on historical instrumentalobservations of the weather in the Nether-lands within the framework of the E.C.Climatological Research Programme,HISKLIM (HIStorical CLIMate). Theobjective of HISKLIM is to make histori-cal meteorological observations fromDutch-language sources available in adigitized format (see KNMI, 2009).Weather measurements in the nine-

teenth century were done using self-recording apparatus with a moderatedegree of reliability. Due to changes in thenumber of readings, the time at which the

readings took place, measuring position,measuring instruments, etc., the climatetime series are not homogeneous5. Toincrease their usefulness, the data for dailymean temperature and maximum andminimum temperatures have to behomogenized. Various procedures havebeen developed to calculate the tempera-ture for every hour of a given day from asmall number of at least two regular rea-dings on the same day (Van der Hoeven,1992; Van Engelen & Geurts, 1983). Weused a slightly adapted version of themethod of Van Engelen & Geurts (seeVan Duin, 2008) to calculate homoge-neous mean daily temperature and maxi-mum and minimum temperatures from1854 to 1950. Given the differences inclimatological conditions between thefour provinces we decided not to use onesingle temperature series for all fourregions but to apply region-specific mea-surements. For Zeeland we used datafrom the Vlissingen weather station, forGelderland those of the Utrecht/De Biltweather station, and for Drenthe andGroningen those of the Groningen/Eeldeweather station6.The distance between theweather stations and the areas for whichthe weather measurements were consi-dered indicative was modest. For Zeelandthe largest distance between the weatherstation and any of the province’s munici-palities for which we had mortality datawas 45.9 km, for Gelderland it was 109.2and 106.7 km, and for Drenthe andGroningen 63.0 respectively 56.8 km7.For each of the three stations we calcu-

lated a series of indicators of (extreme)temperature conditions. The average daily24-hour temperature was rather stableover time but varied between theregions/stations, with Zeeland registeringslightly higher temperatures than Dren-the/Groningen and Gelderland but with


66

much more frequent extremes in thelatter two provinces. Both the number oftropical days (maximum temperatureabove 30°C) and the number of ice days(maximum temperature below 0°C) weremuch higher in Drenthe/Groningen andGelderland. Mean heat and cold valuesshowed a comparable pattern. TheNetherlands Royal Meteorological Insti-tute classifies winters and summers byusing annual cold (or Hellmann) values8and heat values9. The Hellmann value wasmuch lower in Zeeland and the meanheat value was highest in Gelderland,followed by Zeeland and Drenthe/Groningen.According to the official definition by

the Netherlands Royal MeteorologicalInstitute, a heat wave is defined as aperiod of at least five days, each with amaximum temperature of at least 25°C(called summer days), including at leastthree days with a maximum temperatureof at least 30°C (called tropical days),measured at the De Bilt station located inthe centre of the Netherlands. Applyingthis definition to the weather stations inour study, there were only four heatwaves in Zeeland in the 1855-1950period against 18 in Drenthe/Groningenand 27 in Gelderland. A cold spell is aconsecutive series of at least five ice days(maximum temperature below 0°C)including at least three days with severefrost (maximum temperature below–10°C), thus according to this definitionthere were eight cold spells in Zeelandagainst 31 in Drenthe/Groningen and 33in Gelderland.Mortality data for the four provinces are

available for the 1812-1950 period. Werecoded ages at death into age groups andoccupation/social class of the deceased,their parents and spouse when applicable.Ages at death were classified into the

following groups: stillbirth, first-yearmortality (age at death less than 1 year),deaths at ages 1-4, 5-19, 20-49, 50-74and ages 75 and older.We classified all occupations of

deceased persons, their spouses andparents in a social class system, based ona recently developed coding schemecalled HISCO (Historical InternationalStandard Classification of Occupations)(Van Leeuwen, Maas, & Miles, 2002).HISCO translates occupational descrip-tions into a common code, compatiblewith the International Labour Organi-zation’s International Standard Classifi-cation of Occupations (ISCO68) scheme.These historical occupational titles wereclassified according to a social classscheme recently proposed by Van dePutte and Miles (2005), known as theSocial Power (SOCPO) scheme. Socialpower is defined as the potential toinfluence one’s “life chances” throughcontrol of (scarce) resources and is basedon economic factors (like self-employ-ment, skill and authority) and culturalresources (non-manual versus manualoccupations, and nobility and prestigetitles). The merging of economic andcultural power dimensions leads to ascheme with five levels10. We denotethese groups as the elite, middle class,skilled workers, semi-skilled workersand unskilled workers. In view of thespecific position occupied by farmers incontemporary social class mortalitystudies, we excluded them from themiddle class and placed them in a sepa-rate category. Given the small numberswe grouped the elite with the middleclass together in one group. Many of thedeceased could not be placed in a cate-gory, as they were not yet or no longereconomically active at the time of theirdeath.


67

Descriptive statistics for the mortalityand temperature variables by 25-yearperiod and province are presented inTable A-1 of the Appendix. The totalnumber of deaths in the database was1,843,301, ranging from 521,450 in the1855-1879 period to 367,243 for 1930-1955. Around 18 per cent of all deathsconcerned infant mortality (below the ageof 1 year), and around 42 per cent personsaged 50 or older. The percentage of infantdeaths was much higher in Zeeland. Thedistribution of deaths by age group showsa shift from younger to older age groups.In the first two periods around 23 percent of all deaths concerned mortalitybelow age 1, dropping in the most recentperiod to less than 8 per cent.The share ofthe oldest age groups, 50 years and older,increased from 29 to almost 64 per cent.The distribution by social class shows thata large majority of the deaths belonged inthe laboring classes—unskilled and semi-skilled workers in and outside agriculturetogether constituted around 33 per centof all deaths, and skilled workers around10 per cent. Nearly 12 per cent of thedeaths happened in farming families.Upper and middle classes made up 11 percent of the total. In all provinces thepercentage of deaths of unskilled andsemi-skilled workers in and outside agri-culture decreased considerably over time.This was also the case with farmers. At thesame time there was, particularly in themost recent period, an enormous increasein the number of deaths with social classunknown (that is, either unknown or nooccupation given, particularly amongwomen).

Method

We use statistical modeling to study thelink between extreme temperatures and

mortality. The approach we adopt issimilar to the one used in two recentstudies of the impact of heat waves andcold spells on mortality in the Nether-lands during the 1979-1997 period(Huynen et al., 2001) and in the Dutchprovince of Zeeland for 1855-2006(Ekamper et al., 2009). With regressionanalysis we can fit the relationshipbetween a dependent variable (dailynumber of deaths) and one or more inde-pendent variables (like daily averagetemperatures, long-term time trend andseasonal pattern). The resulting estimatedregression model describes the relation-ship between the dependent variable andthe independent variables in terms ofregression coefficients. The regressioncoefficients indicate the effect of thesingle independent variable on thedependent variable. Several techniques forcarrying out statistical regression analysishave been developed. Poisson regressionfits models with a dependent variable thatdenotes the number of occurrences(counts) of an event. Since we are dealingwith count data (number of deaths) wethus need to use a Poisson regression.Poisson regression assumes the mean ofthe dependent variable (mean number ofdeaths) to be equal to the variance of thatvariable. However, in our case theobserved variance of the dependent vari-ables (total numbers of deaths, numberof deaths in selected age groups and socialclasses) is generally greater than theirmean. This is known as overdispersion.To account for overdispersion of ourdependent variables we need to use aspecial case of Poisson regression, nega-tive binomial regression (see Cameron &Trivedi, 1998; Hilbe, 2007; McCullagh& Nelder, 1989).In our analyses the daily total

numbers of deaths, as well as the


68

number of deaths in selected age groupsand social classes, were thus related tothe daily average temperatures usingnegative binomial regression models forthe whole dataset (1 January 1855 to 31December 1950) and subsets (25-yearperiods, heat waves and cold spells),controlling for long-term time trendand seasonal pattern11. In the analysesthe winter period includes the coldestmonths in the Netherlands: December,January and February. As many studieshave shown, seasonality is ever-changingover time both in the Netherlands(Kunst, Looman, & Mackenbach,1991) and other countries (see e.g.Eilers, Gampe, Marx, & Rau, 2008;Lerchl, 1998; Marcuzzi & Tasso, 1992;Seretakis et al., 1997). To account forthe varying cyclical seasonal pattern inthe regression models, we used thefollowing general expression of theGampe and Rau (2004) seasonal timeseries modulation model to estimate thelong-term time trend and seasonality forraw counts (see Eilers et al., 2008):log(µt)= υt+ft cos(ωt)+gtsin(ωt)where t=1,…,T and ω=2π/p (where p isthe period, in our analyses the number ofdays per year). The smooth long-termtime trend υt to account for long-termtrends resulting from changes in e.g.population size and structure and socioe-conomic and health care conditions wasincluded as a restricted 7 knots cubicsmoothing spline. The ft and gt parame-ters, describing the local amplitudes of thecosine and sine waves, were included asrestricted 7 knots cubic smoothing splinesof the annual f and g estimates of theGampe-Rau model. The resulting esti-mated varying cyclical seasonal patternover the years is included as one of theindependent variables in the negative

binomial regression model. The varyingcyclical seasonal pattern was estimated forall four provinces separately.Both extremes of temperature have

adverse effects on health, which causescomplications in modeling. Mostresearchers have dealt with this problemby concentrating only upon either coldeffects or heat effects; we prefer tomodel heat and cold simultaneously byusing information on the V-like linkbetween mortality and temperature. Toaccount for this link (see e.g. Huynen etal., 2001), average daily temperatureswithin the model were measured by twocomplementary variables, heat (0 ifaverage temperature was lower than theoptimum value, otherwise averagetemperature minus optimum value) andcold (0 if average temperature washigher than the optimum value, other-wise optimum value minus averagetemperature). The optimum valuecorresponds to the average value of thetemperature with the lowest mortalitylevel found by Huynen et al. for theNetherlands, 16.5°C.Temperature variables were also

constructed in line with Huynen et al.(2001). Lag temperature variables werecalculated by averaging values for heatand cold over lag periods that increasedexponentially in size: lag times 1-2, 3-6,7-14 and 15-30 days. The general formof the regression model used can bedescribed by:log(yi) =β0+β1hi+β2hi-1,i-2+β3hi-3,i-6+β4hi-7,i-14+β5hi-15,i-30+β6 ci+β7ci-1,i-2 +β8ci-3,i-6+β9ci-7,i-14+β10ci-15,i-30+β11siwhere yi (dependent) is the number ofdeaths on day i, hi (heat) is the averagevalue for heat on day i, hi-1,i-2 to hi-15,i-30the average heat values for lag times 1-2to 15-30, ci (cold) the average heat


69

values for cold on day i, ci-1,i-2 to ci-15,i-30the average cold values for lag times 1-2to 15-30, and si (seasonality) thesequential value of the long-termseasonal trend for day i estimatedbeforehand by the Gampe-Rau model.β0…βj are the regression coefficients.Negative binomial regression analyses

were applied to both the average dailytotal number of deaths and the averagedaily number of deaths by sex, age groupin years (<1, 1-4, 5-19, 20-49, 50-74,and 75 or older) and social class(unskilled workers, semi-skilled workers,skilled workers, farmers, and middle classand elite) in years with heat waves andcold spells. Additionally, negative bino-mial regression analyses with respect toaverage daily total number of deaths wereapplied to shorter (25-year interval) timeperiods (1855-1879, 1880-1904, 1905-1929 and 1930-1950) during allsummers and winters per period for totalmortality and for farmers and unskilledworkers.

RESULTS

We start with a descriptive analysis ofthe effect that periods of extreme heatand cold could have on mortality. Fourexamples are given to get an idea ofthese effects for specific years. To thatend we selected four years with extremeheat and four winters with extreme cold.The Netherlands Royal MeteorologicalInstitute classifies winters and summersby using annual cold (or Hellmann)values and heat values calculated fromthe measurements of the Utrecht station(see previous section). To be able toselect years with extreme heat and coldin the same year in all four provinces wecalculated the annual heat and coldvalues for all three stations. However,

the ranking of the years is different perprovince, therefore we selected foursummers and four winters that on aver-age ranked the highest when combiningthe rankings of the three stations12. Weselected the extreme-heat summers of1868 (average ranking 2nd), 1884 (4th),1911 (3rd) and 1947 (1st), and theextreme-cold winters of 1854-55 (2nd),1890-91 (3rd), 1928-29 (4th) and 1946-47 (1st). Table 2 gives some informationabout the selected years. In all stationsthe summer of 1947 had by far thehighest heat value (the Utrecht stationcounting as many as four heat waves),followed by the summer of 1911 (theUtrecht station being an exception).The winter period of 1946-1947 was byfar the coldest at all stations, its coldvalue higher than any other year.Figure 2 shows the daily mortality

ratio (the observed number of deaths ina day divided by the average number ofdeaths per day in that year) and theaverage 24-hour temperature in yearscharacterized by heat waves, whereasFigure 3 does the same for years charac-terized by cold spells. For reasons ofreadability we chose to present the datafor the Gelderland province only.As shown in several studies, the effects

of extreme heat or cold may consist of amore or less instantaneous effect and amore delayed effect. In 1911 for exam-ple, in Gelderland after a first heat wavestarting around July 20 mortality startedto increase after one week. Temperaturepeaked around July 29 and again severaldays during August and September.Average temperatures for lag days 7-14and 15-30 show patterns that are moreor less in line with the mortality pattern.Yet there were very strong differencesbetween provinces and between specificyears in the effect that summers with

Drenthe/Groningen

Utrecht

Vlissingen

Heatw

aves

Heat

Sum.

Trop.

Heatw

aves

Heat

Sum.

Trop.

Heatw

aves

Heat

Sum.

Trop.

Period

Ndays

value

days

days

Ndays

value

days

days

Ndays

value

days

days

1868

00

97.9

252

114

208.2

4311

00

142.9

282

1884

00

22.7

100

00

129.0

320

00

137.8

222

1911

217

123.8

3212

219

135.0

4212

17

166.8

3111

1947

214

180.1

4012

438

221.3

5718

111

179.8

334

Drenthe/Groningen

Utrecht

Vlissingen

Coldspells

Cold

Ice

10°C

Coldspells

Cold

Ice

10°C

Coldspells

Cold

Ice

10°C

Ndays

value

days

days

Ndays

value

days

days

Ndays

value

days

days

1854-55

228

266.2

3815

333

260.7

3318

111

162.1

246

1890-91

128

240.1

457

327

256.2

3612

113

182.3

303

1928-29

111

270.1

3416

00

227.1

2615

111

139.5

245

1946-47

239

405.6

5217

340

342.8

4621

00

220.8

401


70

Tab.2

Num

beran

dD

urat

ion

ofH

eatW

aves

and

Cold

Spells,

Hea

tand

ColdVa

lues,N

umbe

rof

Sum

mer

Day

s,Tr

opical

Day

s,IceD

aysa

ndD

aysw

ithM

inim

umbe

low

-10°

Cby

Wea

ther

Stat

ion

forSe

lected

Sum

mer

sand

Win

ters


71

Fig. 2 Daily Mortality Ratio and Average 24-Hour Temperature (°C) in the Dutch Province of Gelderlandin the Summers of 1868, 1884, 1911 and 1947

(a) Summer 1868

(b) Summer 1884

Mortality ratio = observed number of deaths per day divided by average number of deaths in the sameyear

(c) Summer 1911

(d) Summer 1947

72



73


heat waves had on mortality. In all fourprovinces the summer of 1868 hadimmediate and delayed effects onmortality, with mortality ratios reachingvalues around 2 (that is, a doubling ofthe expected number of deaths). Thesummer of 1884 had a very strong andlong-lasting effect on mortality inZeeland but left the mortality patterns inGelderland, Groningen and Drenthealmost unaffected. The famous summerof 1911 again had an effect mainly onmortality ratios in Zeeland but almost noeffect was observed in the other provinces.Finally, the summer of 1947, the secondhottest since registration of time weatherdata started in the Netherlands, left notrace on the provinces’ mortality ratio.Figure 3 shows mortality and tempe-

rature in the coldest winters (Decemberto February) for 1855-1950. Again, herewe find rather different reactions peryear and province. For example, duringthe winter of 1854-1855 strong effectswere found in Zeeland, Groningen andGelderland with doubled mortalityratios but only a weak reaction followedthe period of extreme cold in Drenthe.During the winter of 1890-1891,mortality ratios in Zeeland againresponded directly to the temperaturedecrease, as did those in Gelderland to alesser degree, but in Drenthe andGroningen no reaction was found what-soever. The winter of 1928-1929 didlead to a strong increase in mortality inGelderland, Groningen and Drenthe,experiencing more than twice thenormal expected numbers of deaths perday. The winter of 1946-1947 did notprovoke outspoken mortality peaks inany of the four provinces.Figures such as those presented above

leave a lot of room for different inter-pretations of the relationship between

temperature and mortality. Theycannot take the effect of seasonalityinto account just like that of above- orbelow-optimal temperatures, nor dothey allow reckoning with the effect oflong-term trends in numbers of deathsor other relevant factors such as thechanging age distribution of deaths.For that reason we now turn to multi-variate statistical models, which offer asolution for at least some of these pro-blems. They allow us to take a real stepforward compared to the mainlydescriptive statistical analyses presentedby the contemporary medical doctorsand statisticians.The results of the regression model

explained in the previous section arepresented in Tables A-2, A-3 and A-4 ofthe Appendix. Table A-2 presents theregression coefficients of the modelapplied to all years with either a heatwave or a cold spell (as defined in theprevious section), Tables A-3 and A-4do the same for all summers andwinters. In Table A-2 the model wasapplied to total mortality (total numberof deaths per day) and to number ofdeaths by sex, age group and social class.All models were calculated with the fullmodel including seasonal time trend.The general relationship between

mortality and temperature can be judgedby the ordinary r² and, in case of overdis-persion, by the overdispersion-based r²13.The model fits the Zeeland data muchbetter than that for Gelderland andGroningen and even more so than thatfor Drenthe. Furthermore, models fitthe temperature-mortality link muchbetter among the elderly than amonginfants, children and adults, butZeeland is an exception to this rule, ashere the models reach their highestvalues of r² for infants.

Fig. 3 Daily Mortality Ratio and Average 24-Hour Temperature (°C) in the Dutch Province ofGelderland in the Winters of 1854-1855, 1890-1891, 1928-1929 and 1946-1947

(a) Winter 1854-55

(b)Winter 1890-91



74

(c)Winter 1928-29

(d)Winter 1946-47



75


76

The regression coefficients for heatwith respect to total mortality show asignificant positive instantaneous effectof heat with a practically equal size in allfour provinces. The 0.0372 effect ofheat for day 0 in Drenthe means that a1°C increase above the optimumtemperature is associated with a 3.79 percent increase in daily number ofdeaths14; for Gelderland, Groningen andZeeland the increases were respectively3.0, 3.6 and 4.1 per cent. A statisticallysignificant delayed effect of heat wasalso observed in Drenthe, Gelderlandand Zeeland for lag days 1-2; withrespectively 1.8, 1.4 and 2.7 per centincreases in daily number of deaths forevery 1°C increase above the optimumtemperature, the size of that effect waslower than the immediate one. Delayedand significant effects were alsoobserved in all four provinces for longerlags, except for the non-significant coef-ficient for lag days 3-6. The effect ofheat for lag-days 7-14 and 15-30 variedfrom 1.4 and 2.3 per cent in Drenthe to1.9 and 3.2 per cent in Gelderland, 1.6to 2.9 per cent in Groningen and 3.1and 8.5 per cent in Zeeland. Duringcold spells the regression models showednegative effects for the immediatetemperature change, and these weresignificant in Drenthe, Groningen andGelderland, where a 1°C decrease belowthe optimum temperature caused a lessthan 1 per cent decrease in daily morta-lity. In Drenthe and Gelderland therewere significant delayed effects oftemperature changes for lag-days 1-2and 7-14, leading to respectively 1.3 and0.8 per cent (Drenthe), 1.2 and 0.5 percent (Groningen) and 1.1 and 1.0percent (Groningen) increases in dailymortality. In general, the effects of heatwere thus much stronger than those of

cold. The pattern appeared to be similarfor men and women in all provinces.For most age groups the links between

temperature and mortality were ratherweak, with the exception of infantmortality (age <1 year). In all provincesimmediate effects of heat were observedin all age groups, in most cases withstronger effects found for infants than forolder age groups. Delayed effects (for days7-14 and 15-30) were consistently foundamong the youngest age groups; for theelderly the situation varied by province.People aged 75 or older suffered fromstrongly increased effects as death ratios inDrenthe and Groningen show, incontrast to Gelderland and Zeeland.Remarkably enough, among infants thedelayed consequences of heat (for lagdays 7-14 and 15-30) were strongerthan the immediate ones.For cold, immediate effects were visi-

ble in Gelderland and Groningenamong infants and children as well asamong the elderly, but the effects werecontrary to what could be expected,with lower-than-optimal temperaturesleading to lower mortality. No imme-diate effects were observed in Drentheand Zeeland. Delayed effects in theexpected direction for lag days 1-2 werevisible for infants and the elderly inGelderland, Groningen and Drenthe, butnot in Zeeland. For longer lags (7 days ormore), the elderly in Drenthe, Groningenand Gelderland were undergoing morta-lity-increasing effects. However, thedelayed effects for cold in lag days 15-30had mortality-decreasing effects, particu-larly for infants. These compensatoryeffects suggest a harvesting effect.Table A-2 also shows that the strength

of the relationship between temperatureand mortality varied by social class. InZeeland, unskilled workers were the


77

only group undergoing an immediateeffect of heat, whereas in Gelderland,Groningen and Drenthe almost allsocial classes underwent a direct effect ofexceptional heat. In these last threeprovinces the effects were felt strongestamong workers, especially the unskilled.Delayed effects of heat were absent inDrenthe. In Gelderland, Groningen andZeeland however delayed effects in lagdays 7-14 and particularly lag-days 15-30 were present in almost all socialclasses and were extremely strong for theunskilled. For example, for unskilledworkers in Gelderland the daily numberof deaths rose by 6.2 per cent for every1°C increase above the optimumtemperature during the 15-30 daysbefore the date of death; in Zeeland thecomparable percentage for the unskilledwas an even 15 per cent, but in othersocial classes increases of 10-12 per centwere observed.Immediate mortality-increasing cold

effects were absent in almost all socialclasses in Drenthe and Zeeland, but inGelderland and Groningen small yetsignificant mortality-decreasing effectswere observed among persons of almostall social classes. The delayed effects inlag days 15-30 had mortality-decrea-sing effects, particularly for unskilledworkers. These compensatory effectsagain suggest a harvesting effect. Therewere however effects in the expecteddirection for lag-days 1-2 and 7-14 inDrenthe and Gelderland, and thesewere mainly present among unskilledworkers.All in all, the link between cold spells

and mortality appeared to be muchweaker than that between heat wavesand mortality. Effects of heat waves werealso much more delayed than those forcold spells.

Table A-3 presents the regression coef-ficients of the regression model appliedto all summers and winters for 1855-1950. Models were estimated separatelyfor four 25-year periods, to analyzechanges in vulnerability to heat and coldover time. As in some of these time pe-riods Zeeland in particular experiencedno or very few heat waves or cold spells,we were only able to study the tempera-ture-mortality connection by includingall summers and winters in the model.The relationships between mortalityand temperature, judged by the ordi-nary r², are more or less the same oreven became stronger over time. Theresults indicate that the immediateeffect of heat in all provinces is signifi-cant and rather strong over the entireperiod, yet the strength of the effect isnot constant over time and a clear trendis not visible either. Whereas in Drentheand Groningen the effect is clearly lowerafter 1930 than in any of the earlierperiods for which data are available, inZeeland and Gelderland the effects areas strong after 1930 as they were in1855-1879.Short-term delayed effects of heat

(days 1-2) were absent in all periods inZeeland, but in Gelderland, Groningenand Drenthe they appeared in mostperiods. Roughly speaking, one couldsay that in these three provinces in themost recent period these effects declinedcompared to earlier periods. Again, thelonger-term delayed effects of heat (lagday 7 and beyond) are rather strong inall provinces. Although the exact timecourse differs slightly in the variousprovinces, it is absolutely clear that thedelayed effect of heat has declined oreven disappeared in the course of time,especially after 1930. This applies to allfour provinces. To give an idea of the


78

consequences, one can compare theregression coefficients between dailymortality and the optimum temperatureminus the average temperatures on lagdays 15-30 for the earliest and last periods:whereas in 1855-1879 a 1°C increaseabove the optimum temperature duringlag days 15-30 implied a 5.5 per cent(Drenthe), 4.1 per cent (Gelderland),5.8 per cent (Groningen) and 14 percent (Zeeland) increase in number ofdeaths per day, the delayed effect of heatdeclined after 1930 to less than 1 percent.The direct effect of cold is negative in

all provinces and periods, implying thattemperatures below the optimal lead toa decline in daily mortality. That effectis significant across the periods andprovinces and is only absent in themost recent period in Zeeland. Resultsmore in line with expectations areobserved for delayed effects. In allprovinces significant mortality-increa-sing effects are found for below-optimaltemperatures during lag days 1-2,disappearing after 1930. For lag days 3-6 and 7-14 mortality-increasing effectsof colder-than-optimal temperaturesare observed as well, but there is noquestion here of a decline of that effectover time. For lag days 15-30 the effectchanged from mortality-decreasing tomortality-increasing.Table A-3 also allows us to find out

whether in the course of time there wasa change in vulnerability to extremetemperatures by social class. To that end,we studied how the effects of tempera-ture affected unskilled workers andfarmers in each time period. Forunskilled workers we observe significantand rather strong immediate effects ofabove-optimal temperatures on dailymortality in all provinces. In Drenthe,

Groningen and Zeeland that effectdisappears after 1930, whereas inGelderland one finds an even strongereffect after 1930 than before. Strong,prolonged delayed effects (lag days 7-14and 15-30) are present in all provincesuntil 1930 but disappear after 1930.For farmers the situation is different.

Only in Gelderland is a direct andalmost unchanged effect of heat obser-vable. Delayed effects for lag-days 7-14and 15-30 are observed for Gelderlandand Zeeland in almost all periods. Heretoo, delayed heat effects are no longervisible after 1930.In all provinces there is a mortality-

reducing immediate effect of cold belowthe optimal temperature for unskilledworkers. Short-term (lag days 1-2)delayed effects of cold in the expecteddirection (mortality-increasing) arefound among unskilled workers anddisappear only after 1900 or even onlyafter 1930. Among farmers, immediateor delayed effects of cold are absent.Table A-4 presents the results of a

regression model in which the changingrelationship between temperature andinfant and old-age mortality is studiedfor four 25-year periods. The meannumber of deaths per day in the firstage group declined from 13.7 to 3.4,whereas the number of deaths amongpeople aged 75 and older increasedfrom 4.5 to 12.8. Infant mortalityunderwent strong immediate effects ofheat in every province and almost everyperiod. It is clear that the immediateeffect of heat on mortality in theyoungest age group decreases almostcontinuously over time in all fourprovinces; after 1930 it is only seen inGelderland. Similar tendencies are visi-ble for delayed effects of heat. For lagdays 7-14 as well as 15-30 the very


79

strong delayed effects that were visiblefrom the beginning of the perioddeclined everywhere, although theyremained present even after 1930 inDrenthe and Gelderland. Whereas in1855-1879 a 1°C increase above theoptimum temperature during lag days15-30 implied a 14.6 per cent (Drenthe),10.7 per cent (Gelderland), 7.3 per cent(Groningen) and 22.7 per cent(Zeeland) increase in number of deathsper day, after 1930 the delayed effect ofheat declined to 10.4 per cent (Drenthe),9.1 per cent (Gelderland) , 3.4 per cent(Groningen) and 5.2 per cent(Zeeland). For cold, the immediateeffects again were in the unexpecteddirection and were present in allprovinces and periods. Significantmortality-increasing delayed effectswere visible for lag days 1-2 in allprovinces but disappeared after 1930.In the age group 75 years and older

there was a clear immediate effect ofheat in all provinces; over time, thateffect strongly decreased in Drenthe butremained at more or less the same levelin the other provinces. Strong delayedeffects for lag days 7-14 and 15-30 werepresent in Drenthe but absent in otherprovinces. Remarkably enough, coldhad direct mortality-decreasing effectsfor the elderly, which was visible in allprovinces. Short-term delayed effectswere observed until 1930, whereas inZeeland, Groningen and Gelderlandlonger-term delayed effects (lag days 7-14 and 15-30) were also observed. Aclear time trend was not visible though.From several of the tables we observe a

mortality-decreasing effect for lag days15-30, suggesting a compensatoryharvesting effect, particularly in theearlier periods. Although a thoroughanalysis of harvesting effects in the

longer run falls beyond the scope of thisstudy, exploratory analysis of mortality-decreasing effects in the longer runindeed suggests a harvesting effect afterseveral months, particularly with respectto heat in the earlier periods. Compa-ring monthly mortality rates of theselected heat waves and cold spells(including an extended time intervalafter each) with the monthly mortalityrates of preceding and subsequent years(not including heat waves and coldspells) indicates some harvesting effectsfor heat waves after 4 to 6 months andfor cold after 6 to 8 months. From ahealth perspective heat waves and coldspells do affect life expectancy by redu-cing the number of years of life, but thehigher the harvesting effect the lowerthe number of years of life lost.

CONCLUSION AND DISCUSSION

Our analysis is to our knowledge thefirst one in which the relationshipbetween temperature and mortality wastested for a long historical period withrigorous statistical methods using ratherrefined temperature and mortality data.Our results might have been affected bythe fact that our weather stations were atsome distance from the area for whichwe had mortality data, but in general wethink that that distance was not so farthat it made our results untrustworthy.One might also question whether thechoice of optimum temperature was thebest possible; it might be the case thatthat optimum was different in theperiod that we studied or that it variedby region or age group. The range ofoptimal temperatures used so far inpresent-day studies is so small that wedo not think this would lead tocompletely different results.


80

Our conclusions can be summarizedas follows. Our study showed thatbetween 1855 and 1950 total mortalityunderwent an immediate increase whentemperature rose above the optimalvalue, the size of that effect being moreor less the same in all four provinces. Wealso observed higher mortality related toincreases in temperature 1-2 days beforethe day of death, and strong delayedeffects for lag days 7-14 and 15-30. Thispattern was observed for men andwomen. Effects of heat were strongestfor infants (mortality below age 1) anddelayed effects (for days 7-14 and 15-30) were consistently found among theyoungest age groups too. Immediateeffects of cold were contrary to whatcould be expected, with lower-than-optimal temperatures leading to lowermortality. In general one could say thatthe immediate effects of heat were feltmore strongly among unskilled workers,whereas delayed effects in lag days 7-14and 15-30 were extremely strong forunskilled workers too. All in all, the rela-tionship between cold spells and morta-lity appeared to be much weaker thanthat between heat waves and mortality.Effects of heat waves were also muchmore delayed than those for cold spells.The immediate effects of heat were

not constant over time but no cleartrend was visible. Short-term delayedeffects of heat as well as longer-termdelayed effects declined from 1900 or1930 on. Temperatures below the opti-mum during lag days 1-2 increasedmortality, but this effect disappearedafter 1930 too. Vulnerability ofunskilled workers to heat declined after1930, for immediate as well as longer-time delayed effects. For the youngestage group a decline over time in thestrength of the immediate effect of heat

on mortality is visible, and similartendencies were found for the longer-term delayed effects of heat.The present study corroborates in a

rigorous way many of the conclusionsmade by contemporaries based on muchsimpler methods about the relationshipbetween temperature and mortality. Forexample, our study confirms theoutcomes of studies of early-twentieth-century doctors such as Heynsius vanden Berg, De Lange and others (DeLange, 1913; Gezondheidscommissie 's-Gravenhage, 1913; Heynsius van denBerg, 1912) that heat (and to a lesserdegree cold) had direct and laggedeffects. The fact that we could use dailytemperature and mortality data allowedus to specify the lag periods that had thestrongest effect on mortality. We showedthat these lag effects were different forheat than for cold spells, and contrary tothe situation nowadays, the effects ofextreme heat were stronger for longerlag periods.We also found that children were by

far the most vulnerable group whentemperature reached extremely high orlow values. Whereas present-day studiesalways find a strong effect of extremetemperatures on the mortality levels ofthe elderly, this was hardly visible in ourdata set. Our study also confirmed thatthe lowest social class was the mostvulnerable one during temperature fluc-tuations. The strongest direct anddelayed effects of heat were found forunskilled workers. We also observedstrong regional differences in the effectof heat on mortality—in particular,Zeeland endured strongly increaseddeath rates during extreme weatherconditions, a result that is in line withobservations of contemporaries (Saltet& Falkenburg, 1907; Wybrands, 1914).


81

The few studies that describe long-term changes in the temperature-mortality link have focused on thereduced effects of cold and havedescribed these changes as a conse-quence of diminished exposure thanksto improved housing and workingconditions, clothing and footwear, andtransportation. The elderly were morelikely than other groups to live in homeswith insufficient heating and may havebeen reluctant to turn on the heatbecause of the additional, perhaps unaf-fordable expense involved. Theireconomic and housing situation hasimproved considerably. Indoor coldexposure (a result of poor housingconditions, inadequate heating andlarge temperature differences betweenrooms) and exposure to cold duringbrief excursions outdoors have beenespecially reduced. Cold-related morta-lity was partly due to increased air pollu-tion with SO2 at a time when homeheating was done mainly with coal. Thetransition to other heating methods hasplayed a role in the decrease of excesswinter mortality. Also, delayed effects ofwinter cold such as depleted fuelsupplies and deterioration of the qualityof food became less powerful (Kunst etal., 1991).Studies that observed decreases in the

effect of heat on mortality are ratherrare and do not specifically refer toinfants, the age group that underwentthe strongest change (Carson et al.,2006; Hare, Moran, & Macfarlane,1981). Given the prominent role thatinfant deaths played in the tempera-ture-mortality relationship it is clearthat the causes of changes over time inthe effect of heat as well as the reasonsbehind the higher vulnerability ofunskilled workers and the regional

differences in vulnerability should beprimarily sought in the series of factorsaffecting infant mortality.The strong effects of heat that were

observed among infants—effects thatare no longer visible nowadays—werecaused mainly by high rates of gastroin-testinal diseases (Rombouts, 1902, 98,102). In normal years, mortality due to“diarrhea and enteritis” and other acutegastrointestinal conditions was alreadycharacterized by a strong summer peak,even more so when temperatures wereextremely high for a longer period15.High mortality due to gastrointestinaldiseases was first of all a consequence ofhigh proportions of artificially fed chil-dren. Huck (1997) even argued thatdecreased incidence and duration ofbreastfeeding and the supplementationof breast milk with cow’s milk and otherfoods were the main reasons that thewinter peak in infant mortality (Januaryto March) in English towns changedinto a summer peak in the nineteenthcentury (see also Wybrands, 1914, 91,for a comparable trend in Hamburg).The quality of foods such as milk andbread porridge deteriorated at hightemperatures; the quality of water, usedto dilute milk or prepare other foods,was extremely bad as well during periodsof heat and drought; and purity of fee-ding bottles and teats could not be guar-anteed. In poor homes there was no coolplace to keep either condensed or freshmilk in the summer months (Fildes,1998). Over time, however, infantsbecame less and less sensitive to tempe-rature fluctuations: increased frequencyand duration of breast-feeding,increased use of proprietary artificialfood, improvements in the quality offeeding bottles and of drinking waterand milk, etc. have contributed to this


82

change. Insects, responsible for thetransmission of gastrointestinal disea-ses, have decreased in number with theadvent of modern farming methods,changes in settlement and watermanagement, the construction ofsewers and improved public hygiene(McDowall, 1981).Housing conditions also played a role.

High housing density and sanitationproblems facilitated the spread ofdisease during hot weather, but startingin the first decades of the twentiethcentury the crowding of people indoors(which increases temperatures andhumidity) and cooking in living spaces(which has the same effect) decreased.More and better bedding and clothingfor infants and decreased bed-sharingpositively affected the capacity of infantsto maintain a stable body temperature(Watson, Potter, Gallucci, & Lumley,1998). Changes in food production,storage and distribution were anotherimportant factor. The growth of refri-gerated food storage reduced mortalityfrom many lethal infectious diseasesduring the summer months (Ellis,1972). Food availability (fruits andvegetables) improved and could have apositive effect on weaned and breastfedchildren via their mothers.Most of these factors also played a role

in the excessive vulnerability of unskilledworkers in the past, in the higher vulne-rability of infants in Zeeland and in theimprovement of the position of unskilledworkers over time. Poor unskilled indus-trial and farm workers and their familymembers were tremendously vulnerableto extreme weather circumstances as aconsequence of bad and crowded hou-sing, insufficient clothing and footwear,harsh working conditions and infantmalnutrition (due to low frequency of

breastfeeding and inadequate artificialfeeding). For a long time, infant morta-lity in Zeeland was by far the highest inthe Netherlands, a combined effect oflow incidence of breastfeeding and theatrocious condition of the drinking waterand sanitation (Commissie belast met hetonderzoek naar den toestand derkinderen in fabrieken arbeidende, 1869;Fokker, 1877). The gradual salinizationof surface and ground water provided anideal environment for the larvae of themalaria-carrying mosquito, therebymaking malaria virtually endemic in thispart of the Netherlands until about 1870,especially during summer periods. It wasa mixture of factors which improved thesituation of the lowest social class andthat of the worst-faring province:increased frequency and duration ofbreastfeeding, increased use of pro-prietary artificial food for infants, betterquality of drinking water and milk,improved housing conditions and sanita-tion, changes in food storage facilities,better clothing for infants, heating ofhomes, decreased dwelling density,changes in water management, etc.Cultural, technological and economicchanges were the driving forces behindthis decreased vulnerability to extremeweather conditions.Unfortunately we were not able to

study the specific effects of extremeheat in a real urban environment.Klinenberg (2002, 230-235) hasconvincingly shown how extremeweather conditions might lead to disas-trous consequences for vulnerable resi-dents in contemporary cities due toisolation, pointing to the rise of anaging population of urban residentsliving alone without sources of contactand social support, and extreme socialand economic inequality manifest in

Peter EKAMPER

Netherlands InterdisciplinaryDemographic Institute (NIDI),

P.O. Box 11650,2502 AR The Hague,

[email protected]

Coen VAN DUIN

Statistics Netherlands (CBS),P. O. Box 24500,

2490 HA Den Haag,Netherlands.

[email protected]

Frans VAN POPPELNetherlands Interdisciplinary

Demographic, Institute (NIDI),P.O. Box 11650,

2502 AR The Hague,Netherlands.

[email protected]

Kees MANDEMAKERS

International Institute of Social History,(IISG),

P.O. Box 2169,1000 CD Amsterdam,

[email protected]


83

spatial concentration and social separa-tion of the affluent and the impove-rished. In several aspects the nineteenthcentury city was different from present-day urban areas. It is argued that elitesintermingled freely with lower stratamembers, especially servants, andcould not remain isolated from them.Only starting in the late nineteenthcentury did more affluent groupsdistance themselves from highermortality groups and areas, as the useof servants declined and residentialsegregation increased (Smith, 1991).

Furthermore, there is no doubt that inthe past families were major providersof health care for the weak and theelderly, and this might have led to a lessisolated existence of the latter. In thenear future, death certificates for thelarger cities in the Netherlands willbecome available for the same periodthat is studied here. The “socialautopsy” of temperature disasters thatwill then become possible will teach usa great deal about the living conditionsof the urban population in the nine-teenth and early twentieth centuries.

NOTES

1. We wish to thank the Gelders Archief (provinceof Gelderland), the Drents Archief (province ofDrenthe), the Groninger Archief (province ofGroningen) and the Zeeuws Archief (province ofZeeland) for making their data available to us.

2. High death rates due to winter cold remained atopic in temperate zones of Western Europe asthousands of extra persons died there inextremely cold winters (Analitis et al., 2008;Baccini et al., 2008; Healy, 2003; Keatinge et al.,1997; McMichael et al., 2008).

3. There is also some debate concerning thecomparative impact of minimum, maximum and

average temperatures on mortality (Kalkstein &Davis, 1989).

4. Data for Gelderland presently cover onlymunicipalities alphabetically up to the letter V,and relate to 75 per cent of the total number ofdeaths in the province.

5. Times of observation were not standardized forpre-1880s observations. For the Vlissingen station,for example, temperature (in degrees Celsius) wasmeasured starting December 1, 1854 at 9:00,12:00, and 15:00 hours. Starting October 7, 1855,minimum and maximum temperatures wererecorded. Starting December 1, 1857, temperature


measurements took place at 8:00, 12:00 and 15:00hours, and starting April 1, 1859 at 8:00, 12:00and 14:00 hours, but two years later, on April 1,1861, recording took place at 8:00, 14:00 and20:00 hours.

6. The measurements in all three stations werehomogenized by taking into account the changingposition and the changing times of the day. Themethod estimates a daily temperature (T) patternmodeled by a sine curve (for morning temperaturerise and afternoon cooling) and a negative expo-nential curve (for evening and night cooling) withparameters estimated by Van Engelen and Geurts(1983) (for times of minimum and maximumtemperature, sunrise and sunset times, andevening cooling-down tempo) using two or threereadings (R).

7. The location of the weather stations forGelderland and Drenthe/Groningen changedduring the researched period. In the following werefer to weather stations as if they were located inthe province for which we have mortality data.

8. The Hellmann or cold value is calculated bythe summation of all 24-hour mean temperaturesbelow 0°C over the period 1 November – 31March without the minus sign.

9. The heat value is calculated by the summationof all 24-hour mean temperature number ofdegrees above 18 °C over the period 1 May – 31October.

10. Level 5 includes executives, general policymak-ers, supra-local businessmen, non-manual super-skilled workers and members of the nobility. Level4 includes supervisors of skilled workers, localbusinessmen, manual super-skilled workers andnon-manual skilled persons. In level 3 we findsupervisors of semi-skilled and unskilled workers,and manual skilled workers. Level 2 has the locallyoriented self-employed with a minimal capital andsemi-skilled workers. Level 1 comprises unskilledworkers.

11. The time span covers some periods withexceptionally high mortality, e. g. the Spanish flupandemic of 1918 and World War II (May 1940

to May 1945). However, no temperature mea-surements are available from the Eelde andVlissingen stations for the period October 1944to July 1945. Both these high-mortality periodsand the period with missing temperature datawere excluded from the regression analyses.

12. For instance, the summer of 1921 was thesecond hottest in Zeeland but ranked only ave-rage in the other provinces, therefore it was notselected. On the other hand, the summer of 1884ranked relatively high only in Gelderland andZeeland but still ranked fourth on average andwas thus selected.

13. Since the statistical package we used does notproduce a proper r² measure for negative bino-mial regression models, we use an ordinarysquared multiple correlation coefficient for theobserved dependent variable and estimatedvalues. Cameron and Windmeijer (Cameron &Windmeijer, 1996) already indicated that “R²measures of goodness of fit for count data arerarely, if ever, reported in empirical studies or bystatistical packages”. They do however concludethat “use of any of these measures […] is moreinformative than the current practice of notcomputing an R²”. As a measure of goodness offit, additionally to the ordinary squared multiplecorrelation coefficient r² we also use an overdis-persion-based r² developed for negative bino-mial models (Miaou, 1996; Miaou, Lu, & Lum,1996). r² = 1 - (α / αmax), with αmax estimatedfrom a negative binomial model with a constantterm and overdispersion parameter only. Asmaller overdispersion parameter signifies abetter fit.

14. Calculated as transformation of the regres-sion coefficient using the formula 100 x (e - 1).

15. In Zeeland, for example, these gastrointesti-nal conditions caused a doubling in total numberof deaths in July and August; in the exceptionallyhot summer of 1911 compared to 1910 and1912 they doubled the total death toll in July andincreased the number of deaths in August by afactor of 4 and in September by a factor of 2.5.

84

85


ANALITIS, A., KATSOUYANNI, K., BIGGERI, A.et al. (2008), “Effects of Cold Weather onMortality: Results from 15 EuropeanCities within the PHEWE Project”, Ame-rican Journal of Epidemiology, 168(12),1397-1408.

BACCINI, M., BIGGERI, A., ACCETTA, G. etal. (2008), “Heat Effects on Mortality in15 European Cities”, Epidemiology, 19(5),711-719.

BALLESTER, F., MICHELOZZI, P., INIGUEZ, C.(2003), “Weather, Climate, and PublicHealth”, Journal of Epidemiology andCommunity Health, 57(10), 759-760.

BENGTSSON, T. (2004), “Living Standardsand Economic Stress”, 27-60, in Lifeunder Pressure. Mortality and Living Stan-dards in Europe and Asia, 1700-1900, T.Bengtsson, C. Campbell & J. Z. Lee(eds.), Cambridge (Mass.), The MITPress.

BORRELL, C., MARÍ-DELL’OLMO, M.,RODRIGUEZ-SANZ, M. et al. (2006),“Socioeconomic Position and ExcessMortality during the Heat Wave of 2003in Barcelona”, European Journal ofEpidemiology, 21(9), 633-640.

BRESCHI, M., LIVI-BACCI, M. (1994), « Lemois de naissance comme facteur de surviedes enfants », Annales de DémographieHistorique, 169-185.

CAMERON, A. C., TRIVEDI, P. K. (1998),Regression Analysis of Count Data (Vol.30), Cambridge, Cambridge UniversityPress.

CAMERON, A. C., WINDMEIJER, F. A. G.(1996), “R-Squared Measures for CountData Regression Models with Applica-tions to Health-Care Utilization”, Journalof Business & Economic Statistics, 14(2),209-220.

CARSON, C., HAJAT, S., ARMSTRONG, B. et al.(2006), “Declining Vulnerability to

Temperature-Related Mortality in Londonover the 20th Century”, American Journal ofEpidemiology, 164(1), 77-84.

COMMISSIE BELAST MET HET ONDERZOEK

NAAR DEN TOESTAND DER KINDEREN IN

FABRIEKEN ARBEIDENDE (1869), Rapport derCommissie belast met het onderzoek naar dentoestand der kinderen in fabrieken arbeidende.Vierde aflevering. Resultaten van het onder-zoek der sterfte-statistiek, ’s-Gravenhage, TerAlgemeene Landsdrukkerij.

DE LANGE, C. (1913), “De zuigelingen-sterfte aan voedingsstoornissen en stuipente Amsterdam gedurende het derde kwar-taal van 1911”, Nederlands Tijdschrift voorGeneeskunde, 57(2), 91-114.

DE STATISTIEK VAN DEN LOOP DER BEVOLKING

(1911, 7 November 1911), Nieuwe Rotter-damsche Courant, 9-10.

DONALDSON, G. C., KEATINGE, W. R.(2003), “Cold Related Mortality inEngland and Wales; Influence of SocialClass inWorking and Retired Age Groups”,Journal of Epidemiology and CommunityHealth, 57(10), 790-791.

EILERS, P. H., GAMPE, J., MARX, B. D. et al.(2008), “Modulation Models for SeasonalTime Series and Incidence Tables”, Statis-tics in Medicine, 27(17), 3430-3441.

EKAMPER, P., VAN POPPEL, F., VAN DUIN, C.(2009), “150 Years of Temperature-RelatedExcess Mortality in the Netherlands”,Demographic Research, 21, 385-426.

ELLIS, F. P. (1972), “Mortality from HeatIllness and Heat-Aggravated Illness in theUnited States”, Environmental Research,5(1), 1-58.

FILDES, V. (1998), “Infant Feeding Practicesand Infant Mortality in England,1900–1919”, Continuity and Change,13(2), 251-280.

FOKKER, A. P. (1877), “De volksvoeding inZeeland”, Nederlands Tijdschrift voorGeneeskunde, 21, 195-240.

BIBLIOGRAPHICAL REFERENCES

86


GALLOWAY, P. R. (1985), “Annual Variationsin Deaths by Age, Deaths by Cause, Prices,and Weather in London 1670 to 1830”,Population Studies, 39(3), 487-505.

GALLOWAY, P. R. (1986), “Long-Term Fluc-tuations in Climate and Population in thePreindustrial Era”, Population and Deve-lopment Review, 12(1), 1-24.

GALLOWAY, P. R. (1988), “Basic Patterns inAnnual Variations in Fertility, Nuptiality,Mortality, and Prices in Pre-industrialEurope”, Population Studies, 42(2), 275-303.

GALLOWAY, P. R. (1994), “Secular Changesin the Short-Term Preventive, Positive,and Temperature Checks to PopulationGrowth in Europe, 1460 to 1909”,Climatic Change, 26(1), 3-63.

GAMPE, J., RAU, R. (2004), “Seasonal Varia-tion in Death Counts: P-spline Smoothingin the Presence of Overdispersion”, 405-409, in Statistical Modelling: Proceedings ofthe 19th InternationalWorkshop on StatisticalModelling, Florence (Italy), 4-8 July, 2004,A. Biggeri & E. Dreassi & M. Marchi(eds.), Florence, Firenze University Press.

GEDEPUTEERDE STATEN ZEELAND (1869),Verslag van den toestand der provincieZeeland door Gedeputeerde Staten uitge-bragt aan de Provinciale Staten over het jaar1868, Middelburg, Gedeputeerde Staten.

GEMMELL, I., MCLOONE, P., BODDY, F. A.,(2000), “Seasonal Variation in Mortalityin Scotland”, International Journal ofEpidemiology, 29(2), 274-279.

GEZONDHEIDSCOMMISSIE'S-GRAVENHAGE

(1913), Kindersterfte in verband metvoedingsw ze en sociale omstandigheden, te's-Gravenhage en Scheveningen (onderzoek1908 t/m 1912) en middelen ter verbete-ring, 's-Gravenhage, Gebr. Belinfante.

GILL, J. S., DAVIES, P., GILL, S. K. et al.(1988), “Wind-Chill and the Seasonal Vari-ation of Cerebrovascular Disease”, Journalof Clinical Epidemiology, 41(3), 225-230.

GODEFROI, M. J. (1869), “De kindersterftete 's Hertogenbosch beneden het eerste

levensjaar over de jaren 1858 tot 1867”,Tijdschrift Gezondheidsleer, 3, 97-104.

HAINES, A., KOVATS, R. S., CAMPBELL-LENDRUM, D. et al. (2006), “ClimateChange and Human Health: Impacts,Vulnerability, and Mitigation”, Lancet,367(9528), 2101-2109.

HAJAT, S., KOVATS, R. S., LACHOWYCZ, K.(2007), “Heat-Related and Cold-RelatedDeaths in England and Wales: Who Is atRisk?", Occupational and EnvironmentalMedicine, 64(2), 93-100.

HARE, E. H., MORAN, P. A., MACFARLANE,A. (1981), “The Changing Seasonality ofInfant Deaths in England and Wales1912-78 and its Relation to SeasonalTemperature”, Journal of Epidemiology andCommunity Health, 35(2), 77-82.

HEALY, J. D. (2003), “Excess WinterMortality in Europe: a Cross CountryAnalysis Identifying Key Risk Factors”,Journal of Epidemiology and CommunityHealth, 57(10), 784-789.

HEIJBOER, D., NELLESTIJN, J. (2002),Klimaatatlas van Nederland. De normaalpe-riode 1971-2000, Rijswijk, UitgeverijElmar.

HEYNSIUS VAN DEN BERG, M. R. (1912),“Hoe hebben de zuigelingen in enkeleNederlandsche steden den warme zomervan 1911 doorstaan?”, NederlandschTijdschrift voor Geneeskunde, 56(2), 907-920.

HILBE, J. M. (2007), Negative BinomialRegression, Cambridge, CambridgeUniversity Press.

HUCK, P. (1997), “Shifts in the Seasonalityof Infant Deaths in Nine English Townsduring the 19th Century: A Case forReduced Breast Feeding?”, Explorations inEconomic History, 34(3), 368-386.

HUYNEN, M. M., MARTENS, P., SCHRAM, D.et al. (2001), “The Impact of Heat Wavesand Cold Spells on Mortality Rates in theDutch Population”, Environmental HealthPerspectives, 109(5), 463-470.

87


KAISER, R., LE TERTRE, A., SCHWARTZ, J. etal. (2007), “The Effect of the 1995 HeatWave in Chicago on All-Cause andCause-Specific Mortality”, American Jour-nal of Public Health, 97(Supplement_1),S158-162.

KALKSTEIN, L. S., DAVIS, R. E. (1989),“Weather and Human Mortality: An Eval-uation of Demographic and InterregionalResponses in the United States”, Annals ofthe Association of American Geographers,79(1), 44-64.

KEATINGE, W. R., DONALDSON, G. C.(2004), “The Impact of Global WarmingonHealth andMortality”, Southern MedicalJournal, 97(11), 1093-1099.

KEATINGE, W. R., DONALDSON, G. C.,BUCHER, K. et al. (1997), “Cold Exposureand Winter Mortality from IschaemicHeart Disease, Cerebrovascular Disease,Respiratory Disease, and All Causes inWarm and Cold Regions of Europe. TheEurowinter Group”, Lancet, 349(9062),1341-1346.

KEATINGE, W. R., DONALDSON, G. C.,CORDIOLI, E. et al. (2000), “Heat RelatedMortality in Warm and Cold Regions ofEurope: Observational Study”, BritishMedical Journal, 321(7262), 670-673.

KLINENBERG, E. (2002), Heat Wave: a SocialAutopsy of Disaster in Chicago, Chicago,University of Chicago Press.

KNMI (2009), Daggegevens van het weer inNederland - Download [electronic source].KNMI. Retrieved, 2009, from the WorldWide Web:http://www.knmi.nl/klimatologie/daggegevens/download.htmlKOVATS, R. S., KRISTIE, L. E. (2006), “Heat-waves and Public Health in Europe”, Euro-pean Journal of Public Health, 16(6), 592-599.

KUNST, A. E., GROENHOF, F., MACKEN-BACH, J. P. (1994), “The Associationbetween TwoWindchill Indices and DailyMortality Variation in The Netherlands”,American Journal of Public Health, 84(11),1738-1742.

KUNST, A. E., LOOMAN, C. W., MACKEN-BACH, J. P. (1991), “The Decline in WinterExcess Mortality in The Netherlands”,International Journal of Epidemiology, 20(4),971-977.

LANDERS, J. (1986), “Mortality, Weatherand Prices in London 1675-1825: a Studyof Short-Term Fluctuations”, Journal ofHistorical Geography, 12(4), 347-364.

LERCHL, A. (1998), “Changes in the Season-ality of Mortality in Germany from 1946to 1995: the Role of Temperature”, Inter-national Journal of Biometeorology, 42(2),84-88.

LUTERBACHER, J., DIETRICH, D., XOPLAKI,E. et al. (2004),“European Seasonal andAnnual Temperature Variability, Trends,and Extremes since 1500”, Science,303(5663), 1499-1503.

MARCUZZI, G., TASSO, M. (1992), “Sea-sonality of Death in the Period 1889-1988 in the Val di Scalve (Bergamo Pre-Alps, Lombardia, Italy)”, Human Biology,64(2), 215-222.

MCCULLAGH, P., NELDER, J. A. (1989),Generalized Linear Models, London,Chapman and Hall.

MCDOWALL, M. (1981), “Long TermTrends in Seasonal Mortality”, PopulationTrends, 26(4), 16-19.

MCKEE, C. M. (1989), “Deaths in Winter:Can Britain Learn from Europe?”, Euro-pean Journal of Epidemiology, 5(2), 178-182.

MCMICHAEL, A. J., WILKINSON, P., KOVATS,R. S. et al. (2008), “International Study ofTemperature, Heat and Urban Mortality:the 'ISOTHURM' Project”, InternationalJournal of Epidemiology, 37, 1121-1131.

MIAOU, S.-P. (1996), Measuring the Good-ness-of-Fit of Accident Prediction Models(FHWA-RD-96-040), Washington, D.C.,Federal Highway Administration.

MIAOU, S.-P., LU, A., LUM, H. S. (1996),“Pitfalls of Using R² to Evaluate Good-ness of Fit of Accident Prediction

88


Models”, Transportation Research Record,1542, 6-13.

MICHAELOWA, A. (2001), “The Impact ofShort-Term Climatic Change on Britishand French Agriculture and Population inthe First Half of the 18th Century”, 201-218, in History and Climate. Memories ofthe future?, P. Jones & A. Olgilvie & T.Davis (eds.), New York, Kluwer.

NAUGHTON, M. P., HENDERSON, A.,MIRABELLI, M. C. et al. (2002), “Heat-Related Mortality during a 1999 HeatWave in Chicago”, American Journal ofPreventive Medicine, 22(4), 221-227.

O'NEILL, M. S., ZANOBETTI, A.,SCHWARTZ, J. (2003), “Modifiers of theTemperature and Mortality Association inSeven US Cities”, American Journal ofEpidemiology, 157(12), 1074-1082.

ORIS, M., DEROSAS, R., BRESCHI, M. (2004),“Infant and Child Mortality”, 359-398, inLife under Pressure. Mortality and LivingStandards in Europe and Asia, 1700-1900,T. Bengtsson & C. Campbell & J. Z. Lee(eds.), Cambridge Mass., The MIT Press.

PATTENDEN, S., NIKIFOROV, B., ARMSTRONG,B. G. (2003), “Mortality and Temperaturein Sofia and London”, Journal of Epidemio-logy and Community Health, 57(8), 628-633.

PERNEEL, P. G. M. G. (2000), Het beroeps-journaal van Dr. J.F.Ph. Hers, arts te Oud-Beijerland 1881-1915. Een reconstructievan een plattelandspraktijk omstreeks 1900,Rotterdam, Erasmus Publishing.

POUS KOOLHAAS, C. P. (1869), “De kinder-sterfte in den zomer van 1868”, TijdschriftGezondheidsleer, 13(3), 112-114.

PRIESTER, P. (1998), Geschiedenis van deZeeuwse landbouw circa 1600-1910,Wageningen, AAG Bijdragen.

RAU, R. (2007), Seasonality in HumanMortality: a Demographic Approach, Berlin[et al.], Springer.

REY, G., FOUILLET, A., BESSEMOULIN, P. etal. (2009), “Heat Exposure and Socio-Economic Vulnerability as Synergistic

factors in Heat-Wave-Related Mortality”,European Journal of Epidemiology, 24(9),495-502.

REY, G., FOUILLET, A., JOUGLA, E. et al.(2007), «Vagues de chaleur, fluctuationsordinaires des températures et mortalitéen France depuis 1971 », Population-F,62(3), 533-564.

ROMBOUTS, K. H. (1902), Beschouwingenover het geboorte-en kindersterftecijfer vanNederland gedurende het tijdvak 1875-1899, Harlingen, S.W. Houtsma.

SALTET, R. H., FALKENBURG, P. (1907),Kindersterfte in Nederland (in de jaren 1881-1905) (Vol. 19), Amsterdam, Müller.

SERETAKIS, D., LAGIOU, P., LIPWORTH, L. etal. (1997), “Changing Seasonality ofMortality from Coronary Heart Disease”,Journal of the American Medical Associa-tion, 278(12), 1012-1014.

SMITH, D. S. (1991), “Mortality Differen-tials before the Health Transition. Forum:Fatal Years”, Health Transition Review,1(2), 235-237.

TOULEMON, L., BARBIERI, M. (2008), “TheMortality Impact of the August 2003Heat Wave in France: Investigating the‘Harvesting’ Effect and Other Long-TermConsequences”, Population Studies, 62(1),39-53.

VAN DE PUTTE, B., MILES, A. (2005), “ASocial Classification Scheme for Histori-cal Occupational Data: Partner Selectionand Industrialism in Belgium andEngland, 1800-1918”, Historical Methods,38(2), 61-92.

VAN DER HOEVEN, P. C. T. (1992), Etmaal-temperatuur en dagextremen uit termijn-waarnemingen (Memorandum KD 92-08), De Bilt, KNMI.

VAN DUIN, C. (2008), Berekening historischeetmaaltemperaturen, Unpublished manu-script, Voorburg.

VAN ENGELEN, A. F., GEURTS, H. A. M.(1983), Historische weerkundige waarne-mingen, Deel III: Een rekenmodel dat hetverloop van de temperatuur over een etmaal

89


berekent uit drie termijnmetingen vantemperatuur, De Bilt, KNMI.

VAN LEEUWEN, M. H. D., MAAS, I., MILES,A. (2002), HISCO: Historical Interna-tional Standard Classification of Occupa-tions, Leuven, Leuven University Press.

WATSON, L., POTTER, A., GALLUCCI, R. etal. (1998), “Is Baby too Warm? The Useof Infant Clothing, Bedding and HomeHeating in Victoria, Australia”, EarlyHuman Development, 51(2), 93-107.

WINTLE, M. J. (1985), Aspects of Religion andSociety in the Province of Zeeland Netherlandsin the Nineteenth Century, UnpublishedPh.D., University of Hull, Hull.

WYBRANDS, W. (1914), Over zuigelingen-sterfte, mede naar aanleiding van de zuige-lingensterfte in den Gelderschen achterhoek inde jaren 1903-1907 en 1911, Groningen,Hoitsema.a

SUMMARY

To gain insight into the changing impact ofcold and heat on mortality, we analyzedDutch individual death records in relationto daily temperatures for the 1855-1950period for four of the eleven Dutch provin-ces. By making use of negative binomialregression models we studied whether theeffect of extreme heat and cold varied byprovince, age, sex and social class, and analy-zed the changes in vulnerability to tempera-ture fluctuations. Our study showed thatbetween 1855 and 1950 total mortalityunderwent an immediate increase whentemperature rose above the optimal value,the size of that effect being more or less the

same in all four provinces. We observedincreases in mortality related to temperatureincreases 1-2 days before the day of death,and strong delayed effects for lag days 7-14and 15-30. The immediate and delayedeffects of heat were strongest for infants.Immediate effects of cold were contrary towhat could be expected. The immediate anddelayed effects of heat were felt the strongestamong unskilled workers. Short-termdelayed effects of heat as well as longer-termdelayed effects declined from 1900 and1930 on. The vulnerability of unskilledworkers and infants to heat declined after1930.

90


Le changement à long terme des relations entreles variations temporelles de la mortalité etcelles des températures offre une excellenteoccasion d’approfondir notre connaissance desconséquences du changement climatiqueglobal pour la santé publique. La connaissancede ces changements a en même temps une trèsgrande importance pour les historiens parcequ’elle fournit des informations pertinentes surl’évolution du degré de vulnérabilité de lapopulation face aux chocs externes tels que lesvagues de chaleur ou de froid.Pour analyser cette association, nous avonsutilisé plus de 1,8 million de certificats de décèsprovenant de quatre des onze provinces néer-landaises et des données sur les températuresmoyennes quotidiennes de trois stationsmétéorologiques pour la période 1855-1950.Le nombre quotidien des décès a été modélisé àl’aide d’un modèle de régression binomialenégative avec des décalages du jour 1 jusqu’aujour 30. Le modèle nous permet d’estimersimultanément les effets de périodes de froid etde chaleur, car les températures sont mesuréescomme des déviations d’une température opti-male pour laquelle la mortalité atteint son plusbas niveau. Nous avons adopté successivementdeux approches pour analyser les relations entrela chaleur et le froid intenses et la mortalité :une première approche consiste à s’intéresser àdes années caractérisées par des vagues de

chaleur ou froid ; une seconde approche consis-te à analyser l’association entre les températureset la mortalité pour tous les étés et hivers.Notre premier objectif était de découvrir si leseffets des fluctuations de température varientpar province, âge, sexe et classe sociale, et devoir si, à long terme, les changements desconditions de vie (qu’il s’agisse du travail, dulogement [densité, chauffage], de l’alimenta-tion, des transport, des vêtements, etc.) ontdiminué la vulnérabilité des divers groupes.Notre analyse a démontré qu’entre 1855 et1950 la mortalité totale a connu un accroisse-ment immédiat lorsque la température grim-pait au dessus de la température optimale. Ceteffet était presque le même dans toutes les ré-gions. On a observé en particulier des éléva-tions de la mortalité 1-2 jours après le début dela hausse de la température et des effets trèsforts 7-14 et 15-30 jours après. Chez lesenfants, autant les effets immédiats que leseffets retardés étaient les plus forts. Les ouvriersnon qualifiés ont subi plus que d’autres groupesles effets immédiats et retardés de la chaleur. Aucours de la période, on observe une diminutiondes effets immédiats et retardés de la chaleur, enparticulier dès les années 1900 et 1930. C’estparticulièrement la vulnérabilité des enfants etdes ouvriers non qualifiés par rapport à lachaleur qui s’est affaiblie après 1930.

RÉSUMÉ

91


Political cartoon from Johan Braakensiek (1858-1940) in De Amsterdammer, December 21, 1890

“Whereas the rich enjoyed themselves on ice-skating rinks, in the houses of the day laborers fuel waslacking and the streaming cold kept creeping through roofs, cracks, and slits, causing the death ofinfants”.Herman de Man (1898-1946), De barre winter van negentig (The barren winter of the nineties), Baarn:Bosch & Keunig,‘1936.Text below: Remember: “Charity according to one’s means”.Text upper left: Ice sorrowText upper right: Ice enjoyment

92


Tab.A-1

Cha

racter

isticso

fthe

Mor

talit

yan

dTe

mpe

ratu

reD

ata

inth

eD

utch

Prov

ince

sofD

rent

he,G

elde

rlan

d,G

roni

ngen

and

Zee

land

byPe

riod

,185

5-19

50

aTemperature

measurementsfrom

Groningen

(1855-1905)andEelde

(1906-1950)stations

bTemperature

measurementsfrom

Utrecht(1855-1896)andDeBilt

(1897-1950)stations

cTemperature

measurementsfrom

Vlissingen

station

dNotemperaturemeasurementsfrom

October1944

toJuly1945

eThe

Hellmann(orcold)valueisthesummationofall24hourmeantemperaturesbelow0°Covertheperiod

1Novem

ber-31

March

withoutminussign;the

heatvalueisthesummationofall

24hourmeantemperaturenumberofdegreesabove18

°Covertheperiod

1May-31

October

fAheatwaveisaconsecutiveseriesofatleast5

summerdays(maximum

temperature

25°C)includingatleast3

tropicaldays(maximum

temperature

30°C);acoldspellisaconsecutiveseriesof

atleast5

icedays(maximum

temperature<0°C)includingatleast3

dayswithseverefrost(maximum

temperature<-10°C).

93


Tab.A-2aRe

gressio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lreg

ressi

onbe

twee

nD

aily

Mor

talit

yan

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Seas

onal

Mor

talit

yTr

end

inth

eD

utch

Prov

ince

ofD

rent

hedu

ring

Hea

tWav

esan

dCold

Spells

inth

ePe

riod

1855

-195

0a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945

bAllmodelscalculatedwithfullmodelincludingseasonaltimetrendandtemperatureduring

differenttimelags;m

id/high=middleclass+elite

*Significant(p<0.05),**Significant(p<0.01)

α=overdispersion

parameter,r²=ordinarycorrelationcoefficientbetweenobservedandestimateddependentvariable,overdispersion

basedr²

α=1-(α

/αmax),with

αmaxestimatedfrom

modelwith

aconstantterm

andoverdispersion

parameteronly(seeMiaou,1996).

94


Tab.A-2b

Regr

essio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

Mor

talit

yan

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Seas

onal

Mor

talit

yTr

end

inth

eD

utch

Prov

ince

ofGelde

rlan

ddu

ring

Hea

tWav

esan

dCold

Spells

inth

ePe

riod

1855

-195

0a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags;m



α=overdispersion


basedr²

α=

1-(α

/αmax),with

αmaxestimatedfrom

model

withaconstantterm

andoverdispersion


95


Tab.A-2cRe

gressio

nCoe

fficien

tsof

Neg

ativ

ebi

nom

ialR

egressi

onbe

twee

nD

aily

Mor

talit

yan

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Seas

o-na

lMor

talit

yTr

end

inth

eD

utch

Prov

ince

ofG

roni

ngen

during

Hea

tWav

esan

dco

ldSp

ells

inth

ePe

riod

1855

-195

0a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags;m



=overdispersion

parameter,r²=ordinarycorrelationcoefficientbetweenobservedandestimateddependentvariable,overdispersionbasedr²

α=1-(

α/α

max),with

αmaxestimatedfrom

modelwith

aconstantterm

andoverdispersion


96


Tab.A-2d

Regr

essio

nCoe

fficien

tsof

Neg

ativ

ebi

nom

ialR

egressi

onbe

twee

nD

aily

Mor

talit

yan

dTe

mpe

ratu

rewith

Diff

eren

ttim

eLa

gsCon

trolled

forLo

ng-T

erm

Seas

onal

Mor

talit

yTr

end

inth

eD

utch

Prov

ince

ofZee

land

during

Hea

tWav

esan

dCold

Spells

inth

ePe

riod

1855

-195

0a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags;m



α=overdispersion


basedr²

α=1-(α

/αmax),withα

maxestimatedfrom

modelwitha

constantterm

andoverdispersion


97


Tab.A-3aRe

gressio

nco

effic

ient

sofN

egat

iveBin

omia

lReg

ressi

onbe

twee

nD

aily

tota

lMor

talit

y,M

orta

lityof

Uns

kille

dW

orke

rs,M

orta

lityof

Farm

ersa

ndTe

mpe

ratu

rewith

Diff

eren

tTim

ela

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

onin

theD

utch

Prov

ince

ofD

rent

hedu

ring

allS

umm

ersa

ndW

inters

perPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion


basedr²

α=1-(

α/α

max),with

αmaxestimatedfrom

modelwith

aconstantterm

andoverdispersion


98


Tab.A-3b

Regr

essio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

Tota

lMor

talit

y,M

orta

lityof

Uns

kille

dW

orke

rs,M

orta

lityof

Farm

ersa

ndTe

mpe

ra-

ture

with

Diff

eren

tTim

ela

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

onin

theD

utch

Prov

ince

ofGelde

rlan

ddu

ring

allS

umm

ersa

ndW

inters

perPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion


basedr²

α=1-(α

/α

max),with

αmaxestimatedfrom

modelwith

aconstantterm

andoverdispersion


99


Tab.A-3cRe

gressio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

Tota

lMor

talit

y,m

orta

lityof

Uns

kille

dwor

kers,M

orta

lityof

Farm

ersa

ndTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

onin

theD

utch

Prov

ince

ofG

roni

ngen

during

allS

umm

ersa

ndW

inters

perPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion


basedr²

α=1-(α

/αmax),with

αmaxestimatedfrom

modelwith

aconstantterm

andoverdispersion


100


Tab.A-3d

Regr

essio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

Tota

lMor

talit

y,M

orta

lityof

Uns

kille

dW

orke

rs,m

orta

lityof

Farm

ersa

ndTe

mpe

ratu

rewith

Diff

eren

tTim

ela

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

onin

theD

utch

Prov

ince

ofZee

land

during

allS

umm

ersa

ndW

inters

perPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion

parameter,r²=ordinarycorrelationcoefficientbetweenobserved

andestimated

dependentvariable,overdispersionbasedr²

α=1-(α

/α

max),with

αmaxestimated

from

model

withaconstantterm

andoverdispersion


101


Tab.A-4aRe

gressio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

infa

ntMor

talit

yan

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

ondu

ring

allS

umm

ersa

ndW

inters

perPr

ovin

cean

dPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion

parameter,r²=ordinarycorrelationcoefficientbetweenobserved

andestimated

dependentvariable,overdispersionbasedr²

α=1-(α

/αmax),with

αmaxestimated

from

model

withaconstantterm

andoverdispersion


102


Tab.A-4a(continued).

Regr

essio

nCoe

fficien

tsof

Neg

ativ

ebi

nom

ialR

egressi

onbe

twee

nD

aily

infa

ntM

orta

lityan

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

ondu

ring

allS

umm

ersa

ndW

inters

perPr

ovin

cean

dPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


αv=overdisperionparameter,r²=ordinarycorrelationcoefficientbetweenobservedandestimateddependentvariable,overdispersion

basedr²

α=1-(

α/α

max),withα

maxestimatedfrom

modelwith

aconstantterm

andoverdispersion


103


Tab.A-4b

Regr

essio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

Mor

talit

yof

thePo

pula

tion

Age

d75

+an

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Mor

talit

ytren

dan

dSe

ason

during

allS

umm

ersa

ndW

inters

perPr

ovin

cean

dPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion


basedr²

α=1-(α

/αmax),with

αmaxestimatedfrom

modelwith

aconstantterm

andoverdispersion


104


Tab.A-4b(continued)Re

gressio

nCoe

fficien

tsof

Neg

ativ

eBin

omia

lReg

ressi

onbe

twee

nD

aily

Mor

talit

yof

thePo

pula

tion

Age

d75

+an

dTe

mpe

ratu

rewith

Diff

eren

tTim

eLa

gsCon

trolled

forLo

ng-T

erm

Mor

talit

yTr

end

and

Seas

ondu

ring

allS

umm

ersa

ndW

inters

perPr

ovin

cean

dPe

riod

a

aExceptO

ctober-Novem

ber1918,M

ay1940-July1945


differenttimelags


α=overdispersion

parameter,r²=ordinarycorrelationcoefficientbetweenobservedandestimateddependentvariable,overdispersionbasedr²

α=1-(α

/α

max),with

αmaxestimatedfrom

model

withaconstantterm

andoverdispersion


heatwavesandcoldspellsand …members.ziggo.nl › ekamper › publ › adh-2010-02-ekamper.pdf ·...

Documents