Children and Youth Services Review 54 (2015) 30–40

Contents lists available at ScienceDirect

Children and Youth Services Review

j ourna l homepage: www.e lsev ie r .com/ locate /ch i ldyouth

Case manager job strain in public child welfare agencies: Job demandsand job control's additive effects, and instrumental feedback'smediating role

Mark S. Preston ⁎School of Social Work, Columbia University, 1255 Amsterdam Ave, New York, NY 10025, United States

⁎ Tel.: +1 212 851 2240.E-mail address: mp2557@columbia.edu.

http://dx.doi.org/10.1016/j.childyouth.2015.04.0100190-7409/© 2015 Elsevier Ltd. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 11 February 2015Received in revised form 15 April 2015Accepted 16 April 2015Available online 28 April 2015

Keywords:Child welfareHuman servicesInstrumental feedbackJob strainJob demands–control

Public childwelfare agencies are universally acknowledged as highly stressful work environments. Organization-al and occupational health scholars assert that reducing employee strain perceptions in challenging and strenu-ous workplace settings necessitates control over one's job. Consistent with this idea, the job demands–control(JD–C) model's additive hypothesis states that perceived job demands and perceived job control jointly impactperceptions of job strain. Over three decades of empirical testing, however, has yielded inconsistent findings.This study sought to clarify mixed research results using a sample of 349 public child welfare case managers.Specifically, self-report instrumental feedback was introduced as a possible mediator of the association betweenperceived job control and perceived job strain. In line with the literatures on indeterminate human service tech-nologies and dynamic complex environments, two types of mediational (structural equation modeling andbootstrapping) analyses confirmed the construct's role as an intervening variable when job demands wereperceived as challenging. Data are the first to uncover this mediated relationship within a JD–C framework.More importantly, data call into question the predictive validity and practice utility of the model's seminaladditive hypothesis in public child welfare agencies. Practice implications for public child welfare case managersand ideas for advancing JD–C research are also presented.

© 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Public childwelfare agencies are universally acknowledged as highlystressful occupational environments. In response to this circumstance,child welfare researchers have focused considerable scientific effort to-ward uncovering salient individual – (e.g., resilience, self-efficacy) andorganizational – (e.g., organizational culture and climate) level factorsthat reduce the strain perceptions of public child welfare casemanagers(Pecora, Whittaker, Maluccio, & Barth, 2012). Interestingly, characteris-tics of the job, which directly link employees to their larger organiza-tional context, have received far less empirical study (Preston, 2013a,b). Organizational and occupational health scholars propose a dynamicand interdependent relationship between an employee's perception ofher or his job characteristics and level of perceived job strain (Karasek& Theorell, 1990; Luchman & González-Morales, 2013). Within thisbroad interdisciplinary literature, one particular sub-field, occupationalhealth psychology, has sought to uncover job characteristics thatmollifyperceptions of job strain under mentally challenging and emotionally

demanding workplace conditions similar to those found in the field ofchild welfare (De Lange, Taris, Kompier, Houtman, & Bongers, 2003;Taris & Kompier, 2005a,b).

Empirical evidence from the occupational health psychology litera-ture possesses, at least, two unifying features. First, Karasek's (1979;Karasek & Theorell, 1990) job demands–control (JD–C) model servesas the literature's dominant theoretical and conceptual framework. Sec-ond, extensive cross-sectional, longitudinal, and experimental researchexamining perceived job demands and perceived job control's additiveeffects on various measures of perceived job strain have yielded incon-clusive results (De Lange et al., 2003; Häusser, Mojzisch, Niesel, &Schulz-Hardt, 2010; Van der Doef & Maes, 1999). Inconsistent findingshave led some organizational and occupational health researchers toadvocate for more scientific studies that explore perceived job control'sindirect effects (Terry & Jimmieson, 1999). Further, because the JD–Cmodelwas originally conceived for industrial occupations that used pre-dictable and reliable organizational technologies (Karasek & Theorell,1990), other researchers (Marshall, Barnett, & Sayer, 1997; Pousette,Jacobsson, Thylefors, & Hwang, 2003; Söderfeldt et al., 1996) havequestioned the model's predictive validity and practice utility inhuman service occupations (e.g., child welfare) that employ indetermi-nate technologies.

31M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

In their comprehensive meta-analytic review of feedback interven-tions, Kluger and DeNisi (1996) argue that feedback is a construct thatcuts across and unites seemingly disparate social science theories. Feed-back information, for example, is central to the major theories that un-derpin Karasek's (Karasek, 1979; Karasek & Theorell, 1990) JD–Cmodel.1 Osman (2010) further identifies goal-related or instrumentalfeedback, due to its favorable impact on employee performance indynamic and complex environments (e.g., public child welfare agen-cies), as a contextual factor that affects how individuals interpret thecontrollability of their immediate social surroundings. Thus, the aim ofthis study is twofold. First, this study introduces self-report instrumen-tal feedback as a potential intervening variable within the theoreticallogic and conceptual framework of Karasek's JD–C model. Second, thisstudy empirically tests the construct's mediating role on the control-strain association using a sample of public child welfare case managers.

2. Job demands–control model

The JD–C model (Karasek, 1979; Karasek & Theorell, 1990) hypoth-esizes that perceived job demands and perceived job control jointly im-pact perceptions of job strain through a causal mechanism Karaseklabels active learning (i.e., new knowledge, job skills, and problem solv-ing strategies). Job duties and responsibilities construed as challengingheighten cognitive arousal that employees invest toward confrontingmore demanding performance requirements. If demands of the jobare perceived as too taxing, job performance deteriorates as routinejob skills and problem solving strategies become ineffective (Karasek,1998; Karasek & Theorell, 1990). The perceived gap between actualand desired employee performance transforms excess cognitive arousalinto work anxiety. Work anxiety spawns off-task ruminations that ob-struct the effective processing of information essential for learningnew knowledge and mastering new job skills (Warr & Downing,2000), as well as understanding and resolving unfamiliar work-relatedproblems (Bergman et al., 2012; Daniels, Boocock, Glover, Hartley, &Holland, 2009). Decrements in these core facets of active learning in-crease perceived job strain by lowering positive outcome expectationsand feelings of job competence. Hence, when job demands are experi-enced as onerous, work anxiety inhibits active learning which in turnraises an employee's strain perceptions (Karasek, 1979; Karasek &Theorell, 1990).

Control over one's job, however, is predicted to mitigate perceivedjob strainwhen job duties and responsibilities are judged as formidable.Job control expedites the efficient (re)allocation of surplus cognitivearousal toward overcoming non-routine and/or reoccurring work-related problems, and away from off-task ideations that inducelearning-inhibiting work anxiety (Karasek, 1998; Karasek & Theorell,1990). Job control also facilitates experimenting with novel ideas, andtesting unproven job skills and problem-solving strategies (De Jonge,Spoor, Sonnentag, Dormann, & van den Tooren, 2012; Taris &Kompier, 2005a). Consequences of active learning that resolve mean-ingful work-related obstacles and/or produce value-added performanceoutcomes are routinized and incorporated into an employee's existingrepertoire of coping capabilities (Ohly, Sonnentag, & Pluntke, 2006).An expanded range of coping capabilities strengthen information pro-cessing capacity by inhibiting anxiety-inducing ideations (Warr &Downing, 2000). Feelings of job mastery and favorable performanceresults that emerge from more efficacious employee coping minimizeperceptions of job strain (Daniels, Beesley, Wimalasiri, & Cheyne,2013). Thus, when job demands are judged as burdensome (but notoverwhelming), perceptions of control advance active learning and

1 Action [regulation] theory (Frese & Stewart, 1984), general adaptive syndrome theory(Selye, 1950), job characteristic model (Hackman & Oldham, 1980), and learned helpless-ness theory (Seligman, 1975), all explicitly or implicitly incorporate and discuss goal-related feedback information.

decrease work anxiety which in turn attenuates strain perceptions(Karasek, 1998; Karasek & Theorell, 1990).

Job demands–control researchers investigate the hypothesized jointeffects of perceived job demands and perceived job control onemployee-related physical, psychological, and behavioral outcomes bytesting for additive and interactive effects (Häusser et al., 2010). Theformer predicts the presence of two statistically significantmain effects,while the latter predicts a statistically significant demands–control in-teraction (Karasek, 1979). Although the proposed JD–C interaction hasreceived substantially more theoretical and conceptual attention(Häusser et al., 2010; Van der Doef & Maes, 1999), Karasek (1989) in-sists that the JD–C model's seminal insight is perceived job demandsand perceived job control's additive effects on individual-level out-comes. Several comprehensive literature reviews examining additivemodel studies have consistently uncovered mixed empirical support.Only 41% and 36% of the research studies examined by Van der Doefand Maes (1999), and Häusser et al. (2010), for example, fully support-ed Karasek's additive hypothesis when various measures of psycholog-ical well-being were used. Further, De Lange et al. (2003) found fullsupport for only 47% of high-quality longitudinal additivemodel studiesthat used either physical or psychological indicators of job strain. Similarinconsistencies have been uncovered in the social work literature.For instance, in a sample of New York City human service workers,Rafferty, Friend, and Landsbergis (2001) reported support forKarasek's (1979) additive model, while Kim and Stoner (2008) docu-mented null findings based on a sample of California state-registeredsocial workers.

Explanations for inclusive research results include, but are not limit-ed to, the reliance on cross-sectional research; dimensionality ofKarasek's (1979) job control measure; operationalization of his job de-mands construct; possible confounding of Karasek's job demands mea-sure with his job control and job strain measures; incongruencebetween the type of demand employees encounter and type of controlat their disposal; and omitted control variables, such as socioeconomicstatus (De Jonge & Kompier, 1997; Kain & Jex, 2010). While instructive,these research design, psychometric, and conceptual modifications fail toaddress the unique occupational environment the JD–C model was orig-inally intended to confront. In their ground-breaking book,HealthyWork,Karasek and Theorell (1990) state that the JD–C model's theoreticalorientation was purposefully developed for factory-like “work environ-ment[s] where stressors are routinely planned, years in advance” and“these stressors… occur day in and day out for decades” (p. 85–86).Hence, predictable and reliable organizational technologies used inindustrial occupations may constitute the environmental contingencynecessary for fostering perceived job control's strain-reducingproperties. De Jonge and Kompier (1997) and others (Marshall et al.,1997; Pousette et al., 2003; Söderfeldt et al., 1996) have also identifiedand discussed this potential occupational-level boundary condition.

3. Indeterminate organizational technologies

3.1. Industrial occupations

Organizational technologies are purposively designed tools andtechniques that transform an agency's untreated inputs into prescribedoutputs (Hasenfeld, 1983; Sandfort, 2010). Technologies associatedwith industrial occupations are scientifically-based and adopt procedur-al knowledge anchored in tangible cause–effect relations (Austin, 2002;Hasenfeld, 2010). Because these technologies are highly reliable andpredictable, perceived job control minimizes strain perceptions in, atleast, two important ways. First, employees who operate or interfacewith industrial technologies can use control over their job to developempirically-based procedural knowledge that accurately estimates theprobable results of their behavioral actions (Hasenfeld, 1983). Tangibleaction–outcome relations , in other words, reduce workplace ambiguityconcerning the identification, selection, and execution of requisite

32 M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

behavioral actions (Becker, 2004; March & Simon, 1958). Decreased op-erational uncertainty involving the association between behavioral ac-tions, outcomes, and accompanying consequences limit the informationprocessing demands placed on an employee's mental resources (Becker&Knudsen, 2005). Surplus cognitive arousal is channeled toward the rou-tinization and internalization of new knowledge, job skills, and problemsolving strategies (i.e., active learning) designed, in part, to reestablishcoping equilibrium. Additional coping capacity eases work anxietywhich in turn lowers strain (Karasek & Theorell, 1990).

Control over one's job in industrial occupations also mollifies per-ceived job strain by strengthening an employee's belief that preferred or-ganizational technologies will yield quality production results (Becker,2004;March& Simon, 1958). Technologies used in industrial occupationstypically possess a pre-existing menu of preferred response options thatguide current and future behavioral actions. This narrow list ofempirically-based best practices improves the likelihood thatemployeeswill identify, select, and implement response options that gen-erate performance outcomeswith the highest expected utility (i.e., activelearning) (Becker & Knudsen, 2005). Greater operational predictability,due to positive outcome expectations, alleviates perceived job strain byinstilling the belief that preferred response options are masterable andpredetermined production goals are achievable (Karasek, 1979; Karasek& Theorell, 1990). Thus, when demands of the job are challenging, orga-nizational technologies employed in industrial workplace settings reduceperceived job strain by instituting well-structured occupational environ-ments – evidence-based best practices to guide procedural knowledge,specify and clear performance standards that strengthen positive out-come expectations, and transparent and interpretable cause–effect asso-ciations that increase feelings of job competence – where employeescan use control over their job to engage in effective active learning(Becker & Knudsen, 2005; Hasenfeld, 1983; Karasek & Theorell, 1990).

3.2. Child welfare occupational environments

In comparison to industrial occupations, organizational technologiesthat public childwelfare agencies use are described as highly indetermi-nate (Rzepnicki & Johnson, 2005; Smith, 2010). Indeterminate technol-ogies, according to Hasenfeld (1983, 2010), is the distinguishing featureof all human service occupations. Social work researchers have identi-fied several occupational-level characteristics that account for thechild welfare profession's use of and dependence on indeterminate or-ganizational technologies. This partial list includes raw materials,knowledge base, and staff-client relations (Austin, 2002; Hasenfeld,2010). Each of these occupational-level attributes not only contributesto the indeterminate nature of child welfare technologies (Smith,2010), but also influences the level of perceived strain public childwelfare case managers experience (Pecora et al., 2012).

3.2.1. Raw materialsSöderfeldt et al. (1996), posit that job control in industrial occupa-

tions primarily refers to an employee's control over her or his agency'swork methods and workflow processes, whereas perceived control inhuman service occupations, including child welfare, refers to controlover people. Unlike inanimate objects, recipients of public child welfareservices possess personal beliefs and internal motivations that functionwithin highly interdependent cognitive, behavioral, and social systems(Bandura, 1997; Hasenfeld, 1983). Moreover, client thoughts andactions that public child welfare agencies attempt to transform arequite variable and, at times, unpredictable (Smith, 2010). Thus, the inabil-ity to correctly gauge the likely outcome and/or consequences of a partic-ular organizational technology on client attitudes and conduct, relative toa broad set of similar and often indistinguishable technologies, elevatesthe degree of operational uncertainty public child welfare case managersencounter (Darlington, Feeney, & Rixon, 2004; Gambrill, 2008).

Operational uncertainty resulting from indeterminate child welfaretechnologies decreases the probability that public child welfare case

managers will, for example, choose client service interventions highestin expected utility (Rzepnicki & Johnson, 2005). As such, poor client out-come expectations, based on a public child welfare casemanager's priorexperience with or (in)formal knowledge of a specific client service in-tervention or menu of interventions' (in)effectiveness cues off-task ide-ations that raises her or his level of work anxiety. Work anxiety in turnescalates perceived job strain by diminishing a public child welfare casemanager's belief that sanctionable performance standards are attainableor core job requirements are achievable (Preston, 2013a).

3.2.2. Knowledge baseSince unpredictable human beings are the raw material that public

child welfare agencies are legally authorized to transform (Hasenfeld,2010), organizational technologies used by public child welfare agen-cies are rarely based on complete scientific understanding (Littell,2005, 2008). To overcome this occupational-level constraint, publicchild welfare agencies develop practice ideologies (Rapoport, 1960;Smith, 2010). Practice ideologies fill gaps in scientific knowledge withunexamined, and often unconscious, assumptions that diminish opera-tional uncertainty by providing public child welfare casemanagers withcogent rationales for preferred organizational technologies (Hasenfeld,2010). Practice ideologies, however, make it difficult for:

1. public child welfare agencies to produce a baseline of empirical datathat yield a menu of evidence-based practices from which scientifi-cally valid and reliable client service interventions are identified(Littell, 2008), and

2. public child welfare casemanagers to accurately forecast causal con-nections between (un)desirable client outcomes and proceduralknowledge, job skills, and problem solving strategies associatedwith specific client service interventions (Gambrill, 2008;Hasenfeld, 1983).

These by-products of taken-for-granted practice ideologies heightena public child welfare case manager's strain perceptions by elevatingwork anxiety (Preston, 2013b).

3.2.3. Staff–client relationsSince public child welfare agencies are unable to predict with rea-

sonable scientific certainty the unique needs of each family case, theircase managers become the primary tool through which client servicesare identified and, at times, delivered (Smith, 2010). This close recipro-cal staff–client relationship not only confers public child welfare casemanagers with a vast amount of (in)formal authority and decision-making discretion (Sosin, 2010), it also exposes them to an abundantamount of human distress, suffering, and secondary trauma on a fre-quent and repeated basis (Dane, 2000). Given that public child welfarecase managers practice in occupational milieus laden with emotions(DePanfilis & Zlotnik, 2008), indeterminate child welfare technologiesstimulate more than impersonal cognitive calculations on optimal re-sponse options and/or behavioral actions.

Along with agency-reinforced practice ideologies and prior profes-sional experience (Hasenfeld, 2010; Smith, 2010), (un)favorable clientprogress and/or results anticipated from client service interventionstrigger powerful emotional reactions that directly impact the affectivewell-being of front-line staff (Preston, 2013a,b). Unmet agency perfor-mance measures and unfulfilled case plan goals and objectives, for ex-ample, cue off-task musing that increase work anxiety and heightenthe strain perceptions of public child welfare case managers (Kim,2011; Raghunathan & Pham, 1999). In sum, indeterminate public childwelfare technologies produce operationally-uncertain occupational en-vironments – unpredictable child welfare clients, idiosyncratic practiceideologies, and emotional taxing staff–client relations – that increasethe strain perceptions of public child welfare case managers byobstructing their ability to engage in effective active learning, irrespec-tive of control perceptions.

2 The representativeness of thepresent study's sample can bebenchmarked against tworecent national surveys. The NASW (2004) reported the following demographic informa-tion for its members practicing in child welfare: median age, 41 years old; mean job ten-ure, 6 years; respondents self-identified as white, 77%; and respondents self-identifiedas female, 84%. Demographic information reported by Barth, Lloyd, Christ, Chapman, andDickinson (2008) consists of the following: mean job tenure, 7.3 years; respondentsself-identified as white, 67%; respondents self-identified as female, 81%; respondentsself-identified as possessing an undergraduate degree, 48%; and respondents self-identified as having obtained a graduate degree, 30%.

33M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

4. Mediating role of instrumental feedback

Hasenfeld (1983, 2010) states that the perceptions front-line staffhold on their occupational environment are substantially determinedby the technologies their human service agencies employ. Accordingly,public child welfare case managers, due in large part to indeterminatetechnologies, characterize their workplaces as highly uncertain(Gambrill, 2008; Pecora et al., 2012). Organizational scholars assertthat ongoing exposure to operational uncertainty not only necessitatessufficient flexibility over job duties and responsibilities (Jackson, 1989;Preston, 2013a), but also requires an increased flow of information(Aldrich, 2008; Preston, 2013b). Moreover, Galbraith (1973) notesthat the amount of information needed by employees is commensuratewith the level of task uncertainty experienced in their workenvironments.

Experimental and simulation research studies reveal that individualswho encounter difficulty organizing strategic action plans, under condi-tions of high uncertainty, repeatedly monitor their behavioral actions(see Osman, 2010). Drawing on these data-driven studies, Osman(2010) argues that the uncertainty-inducing affect of dynamic and com-plex social environments on purposive human actionwarrants couplingpersonal control with goal-related (i.e., instrumental information) feed-back. Earley, Northcraft, Lee, and Lituchy (1990) define goal-related orinstrumental feedback as information that supports goal attainment. In-strumental feedback is especially critical in occupational settings wherecause–effect relations between behavioral actions, performance out-comes, and their consequences are difficult to establish and evaluate(e.g., operational uncertainty) (Earley et al., 1990).

Karasek (1979; Karasek & Theorell, 1990), as previously noted,posits that ameliorating employee strain perceptions in demanding oc-cupational environments mandates generating new knowledge andcrafting new or improving current job skills. As such, these facets of ac-tive learning represent two causal mechanisms through which self-report instrumental feedbackmaymediate perceived job control's asso-ciation with perceived job strain. Taris and Kompier (2005b) argue thatif employee-related learning is a function of the characteristics of one'sjob, then the receipt of feedback information is the essential ingredient.Further, when social environments are dynamic and complex (e.g., childwelfare), Osman (2010) asserts that the prompt discovery and integra-tion of new knowledge becomes crucial.

Public child welfare case managers who believe that they have suffi-cient control over their jobs can observe the goal-directed behavior ofand/or seek out instrumental information from supervisors, co-workers, and clients (Wielenga-Meijer, Taris, Kompier, & Wigboldus,2010). Novel ideas and insights distilled from process and outcomefeedback promotes the identification and development of creative andinnovative problem-solving strategies (De Jonge et al., 2012; Taris &Kompier, 2005a). Knowledge gains obtained from this facet of activelearning allow public child welfare case managers to anticipate, mini-mize, or prevent atypical work-related problems concerning the estab-lishment case plan goals, implementation and monitoring of case planobjectives, and evaluation of client service interventions (Preston,2013a,b). Heightened performance expectations and feelings of jobmastery, derived from instrumental feedback's uncertainty-reducing ef-fects, in turn mitigate perceived job strain (Karasek & Theorell, 1990).Thus, when job demands are experienced as challenging, perceptionsof job control provide public child welfare case managers with the op-portunity and motivation to learn. Instrumental feedback, on the otherhand, dictates whether or not they do learn.

Skill development is a second causal mechanism through which in-strumental feedback may mediate the control–strain relationship.Frese et al. (1991) state that difficult and uncertain occupational envi-ronments increase the need for employees to update existing skills.Taris and Kompier (2005b) maintain that it is nearly impossible to ex-pand one's job skills without feedback information that causally linksperformance outcomes to behavioral actions. Control over one's job

allows public child welfare case managers to test self-initiated hypoth-eses generated from new ideas and engage in ad hoc experiments onuntested and unfamiliar job skills and/or problem-solving strategies(Taris & Kompier, 2005a,b). Process and outcome feedback garneredfrom these active learning efforts favorably impact job competency be-liefs and performance expectations (Preston, 2007).

Outcome feedback clarifies which client service interventions are (orhave been) themost or least likely to fulfill case plan goals and objectivesand yield meaningful changes in client attitudes and behaviors(i.e., outcome expectations). Process feedback, on the other hand, helpspublic child welfare case managers determine which case plan-relatedjob skills, knowledge gains, and/or problem-solving strategies can be(or have been) mastered or require more fine-tuning (i.e., competencebeliefs) (Earley et al., 1990). Since competency beliefs and outcome ex-pectations serve as the theoretical foundation for perceptions of control(Skinner, 1995), perceived job control should impact perceived jobstrain via self-report instrumental feedback. Thus,without contextual in-formation in the form of process and/or outcome feedback it is impossi-ble for public child welfare case managers to develop new or improvecurrent job skills (Kluger & DeNisi, 1996; Taris & Kompier, 2005a,b).

Indirect support for these arguments is found in both the socialworkand JD–C literatures. In a cross-sectional survey, Pousette et al. (2003)observed that role ambiguity (i.e., task-level uncertainty) indirectlyeffected the level of job satisfaction experienced by human serviceworkers through feedback sign (i.e., positive and negative feedback).Further, experimental work by Jimmieson and Terry (1999) revealedthat under conditions of high task complexity (i.e., operational uncer-tainty), high job demands, and high procedural information(i.e., process feedback), research subjects reported the highest task sat-isfaction under the low, as oppose to high, behavioral control condition.Given the aforementioned theoretical and empirical evidence two hy-potheses were put forth:

Hypothesis 1. Perceived job demands and perceived job control willhave statistically significant additive effects on perceived job strain;such that perceived job demands' main effect will be positive and per-ceived job control’s main effect will be negative.

Hypothesis 2. Self-report instrumental feedback will fully mediate therelationship between perceived job control and perceived job strainwithin a JD–C framework.

5. Method

5.1. Sample characteristics and data collection

This study's sample contained case managers from 11 county-basedpublic childwelfare agencies across the state of NewYork. To ensure ad-equate variability, state counties were purposely chosenwith respect totheir geographic location (i.e., rural, suburban, and urban). Three hun-dred and forty-nine out of 419 usable questionnaires were returnedfor a final response rate of 83%. Demographic data indicated that 87%of survey respondents self-reported as Caucasian and 79% self-reported as female. Fifty-four percent of survey respondents had earnedan undergraduate degree and 51% self-reported as married. Finally, themean age and job tenure, for this particular sample of public child wel-fare case managers, were 41 and 5.3 years, respectively (see Table 1).2

4

Table 1Descriptive statistics, correlation matrix, and reliabilitiesa.

Self-report variable n Mean S.D. 1 2 3 4 5

1. Job strain 349 3.34 0.93 (.84)2. Job control 349 2.67 0.81 − .35 (.80)3. Instrumentalfeedback

349 3.19 1.03 − .36⁎ .27⁎ (.87)

4. Job demands 349 4.31 0.60 .50⁎ − .33⁎ − .14⁎ (.72)5. Age (dichotomous) 349 n/a n/a .16⁎ − .11⁎ − .21⁎ .03 (n/a)

Note. Two-tailed t-test used.a Cronbach's alphas are in parentheses.⁎ p b .05.

34 M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

Copies of the letter of introduction and survey questionnaire, alongwitha self-addressed stamped return envelope, were given to all public childwelfare case managers by each county's staff training and developmentdirector. Included in the letter of introduction were the study's objec-tives, voluntary nature of the research project, and confidentiality safe-guards. In order to increase the study's final response rate, two follow-up contacts were made (Dillman, 2007).

5.2. Measures

5.2.1. Perceived job strainKarasek (1979) defines perceived job strain as psychological strain

that arises from the perceived demands of one's job and perceivedscope of control available to meet those demands. The construct is con-ceptualized as feelings of depression, fatigue, and anxiety (Karasek &Theorell, 1990). Since emotional states are central to work-related psy-chological strain (Warr, 1990), perceived job strain was operationalizedas negative work affect. This outcome variable wasmeasured using fiveitems fromVanKatwyk, Fox, Spector, and Kelloway (2000) 10-item Job-related AffectiveWell-being scale (JAWS) negativework affect subscale.Likert-type response categories ranged from (1) “not at all” to (5) “agreat deal”. Example items included, “I felt depressed while doing myjob?” and “I felt anxious while doing my job?” Previous testing of thismeasure yielded a Cronbach's alpha of .88 (Preston, 2007).

5.2.2. Perceived job controlPerceived job control is defined by Karasek (1979; Karasek &

Theorell, 1990) as potential control over one's tasks and work conductduring the day, and is operationalized as a composite of task or decisionmaking authority and skill discretion. Organizational scholars haveraised numerous psychometric (Terry & Jimmieson, 1999) and concep-tual (De Jonge & Kompier, 1997) concerns with respect to theoperationalization and dimensionality of Karasek's (1979) perceivedjob control measure. As such, several JD–C researchers advocate theuse of a task-specific measure (e.g., Wall, Jackson, Mullarkey, & Parker,1996). Based on this recommendation, this study used four items fromWall et al. (1996)five-itemmultifaceted task-focused controlmeasure.3

Likert-type response categories for thismeasure ranged from (1) “not atall” to (5) “a great deal”. “How much control do you have over whichwork duties to perform in your job?” is an example item. A Cronbach'salpha of .79was observed in a prior test of thismeasure (Preston, 2007).

5.2.3. Self-report instrumental feedbackEarley et al. (1990) define instrumental feedback as information

narrowly-focused on goal-direct action and operationalizes the con-struct as a composite of outcome (e.g., information on goal attainment)and process (e.g., information on effort exerted and/or strategy

3 The question “Howmuch control do you have over the layout of your specificwork ar-ea?”which addresses environmental controlwas omitted. The content validity of this itemwas viewed as questionable for childwelfare casemanagers and the items low factor load-ing (below .40) empirical supported this concern.

effectiveness) feedback.4 Likert response categories for this 4-itemmeasure ranged from (1) “strongly disagree” to (5) “strongly agree”.An example item for process feedback is “In general, I am made awareof how effective my strategies are for completing the work duties ofmy job.” “In general, once specific work duties are completed, I ammade aware of the final results or outcomes.” is an example outcomefeedback item. Previous psychometric testing for thismeasure produceda Cronbach's alpha of .89 (Preston, 2007).

5.2.4. Perceived job demandsJob demands are defined as environmental stressors that employees

perceive as impacting their ability to carry out assigned job duties andresponsibilities (Karasek, 1979). Quantitative and qualitative researchstudies examining the occupational environment of public childwelfareagencies consistently report workload demands as a chronic and prima-ry workplace stressor (Pecora et al., 2012). Perceived job demands,therefore, was operationalized as perceived workload demands. Fouritems for this measure were drawn from Caplan, Cobb, French, VanHarrison, and Pinneau (1975) seven-item quantitative workload scale(α = .71). Prior tests of this abridged job demands measure produceda Cronbach's alpha of .73 (Preston, 2007). Likert-type response catego-ries ranged from (1) “hardly any” to (5) “a great deal”. “How muchwork is expected of you in your job?” is an example item. The meanfor perceived job demands was 4.3 (scale ranged from 1—very little to5—a great deal) which offers empirical evidence that this sample ofpublic child welfare case managers experienced the demands of theirjob as highly challenging.

6. Results

6.1. Descriptive statistics and statistical analyses

Table 1 presents descriptive statistics (number of cases, means, stan-dard deviations), reliabilities, and zero-order bivariate correlations.Prior to examining the two research hypotheses, tests for violations ofOLS regression were performed and none were observed (Hair, Black,Babin, Anderson, & Tatham, 2006).5 Less than 2% of the data were miss-ing; nonetheless, regression imputation was used to address missingvalues. To assess whether data were missing completely at random, di-chotomous coded variables were created for all measures with missingdata (cases with data were coded as 0; cases with missing data werecoded as 1). A bivariate correlational analysis revealed that perceivedjob strain was statistically and significantly related with the dichoto-mous coded age variable (r = .16; p b .05). This finding suggests thatomitting the age variable from the meditational analyses may bias thestudy's final results. Therefore, per Hair et al. (2006) recommendation,the dichotomous coded age variable was included as a control.

Data analyses were carried out in the structural equation modeling(SEM) program AMOS 18.0 and followed Anderson and Gerbing's(1992) two-step process. Step one consisted of estimating a measure-ment model based on the factor structure of the study's four variablesof interest. Step two involved estimating four structural models: an ad-ditive model, a baseline full mediation model, a partial mediationmodel, and a theoretically relevant non-nested alternative mediationmodel. To provide a more stringent test of model fit, the full mediationmodel was compared to the partial mediation and non-nested alterna-tive mediation models (Williams, Vandenberg, & Edwards, 2009).Akaike's information criterion (AIC, Akaike, 1987) was used todetermine the relative fit between the full mediation model and the

Instrumental feedback differs from instrumental support (Greenglass, Fiksenbaum, &Burke, 1996) in that the latter construct encompasses a much broader range of work-related information (e.g., acknowledgment by supervisor that employee appears stressed)and behavioral actions (e.g., supportive listening by co-workers).

5 Skewness and kurtosis (normality); center leverage values, Cook's distance statistic(outliers and nonoutlying influentials); Tolerance and VIF stats (multicollinearity); andWhite's (1980) test (heteroskedasticity).

Table 2AMOS fit measures for the discriminant validity of predictor and criterion measures.

Model χ2 df p-Value CFI RMSEA SRMR

1. Four-factor model (JD, JS, JC, IF) 202.32 113 .001 .959 .049 .0512. Three-factor model (combined JD/JS, JC, IF) 346.99 116 .001 .894 .077 .0683. Three-factor model (combined JD/JC, JS, IF) 441.30 116 .001 .851 .092 .0854. Three-factor model (combined JS/JC, JD, IF) 511.85 116 .001 .818 .101 .0875. Three-factor model (combined JD/IF, JS, JC) 564.79 116 .001 .794 .108 .1356. Three-factor model (combined IF/JC, JD, JS) 581.13 116 .001 .786 .110 .1177. Two-factor model (combined JD/JS/JC, IF) 637.33 118 .001 .762 .115 .0968. Three-factor model (combined JS/IF, JD, JC) 685.24 116 .001 .739 .122 .1039. Two-factor model (combined JD/JS, combined IF/JC) 717.55 118 .001 .725 .124 .12210. Two-factor model (combined JD/JS/IF, JC) 845.27 118 .001 .666 .136 .11611. Two-factor model (combined JD/IF, combined JS/JC) 871.34 118 .001 .654 .139 .15312. Two-factor model (combined JD/JC, combined JS/IF) 902.60 118 .001 .640 .142 .11913. Two-factor model (combined JD/IF/JC, JS) 906.83 118 .001 .638 .142 .14914. Two-factor model (combined JS/IF/JC, JD) 974.42 118 .001 .607 .148 .12415. One-factor model (combined JD/JS/IF/JC) 1120.20 119 .001 .540 .159 .132

Note. JD = perceived job demands. JS = perceived job strain. JC = perceived job control. IF = self-report instrumental feedback.

7 Each model tested included direct paths from the latent variable perceived job de-mands andmanifest variable age (dichotomous) to the latent variable perceived job strain.Further, the latent variable perceived job control was correlated with the latent variableperceived job demands and manifest variable age (dichotomous) in all models.

8 To assess the possible confounding effects of feedback sign child welfare case man-agers were asked how much positive and negative feedback they have received. Thetwo-item feedback signmeasure possessed a Cronbach's alpha of .51 and each item loadedonto a single factor at .82. The feedback signmeasure did not significantly alter the resultsof the meditational analyses and was omitted from the final structural model.

35M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

non-nested alternative mediation model (the lowest AIC value denotessuperiormodel fit).Modelfit for themeasurement and comparisons be-tween other structuralmodels was evaluated using the chi-squared sta-tistic (χ2), comparative fit index (CFI; Bentler, 1990), root mean squareerror of approximation (RMSEA; Steiger, 1990), and standardized rootmean residual (SRMR; Bentler, 1995).6 For sample sizes of less than500, Weston and Gore (2006) suggest that values of .90 or above forthe CFI, .10 or less for the RMSEA, and .10 or less for SRMR are represen-tative of acceptable model fit.

Finally, a bootstrapping test was performed as a means of corrobo-rating the SEM's analyses' full mediation findings (Preacher, Rucker, &Hayes, 2007). Bootstrapping is a re-sampling method that builds pa-rameter estimates based on original sample data. Because the productof two indirect effects creates a skewed distribution, Hayes (2009) ad-vocates this method for testing mediated models. Bootstrapped confi-dence intervals do not assume normality, and as such, are moreaccurate than confidence intervals based on Baron and Kenny's (1986)causal step strategy or Sobel's (1982) test (Hayes, 2009).

6.2. Measurement models

Construct and discriminant validity for this study's measures wasestablished using confirmatory factor analysis with maximum likeli-hood estimation in AMOS 18.0. With respect to construct validity, allitems loaded heavily and uniquely onto their individual latent factorfrom .53 to .83. Table 2 demonstrates that discriminant validity wasestablished between this study's four variables of interest. Given thatstudy datawere collected from single source self-reportmeasures, sam-ple correlations may have been artificially inflated (Podsakoff,MacKenzie, Lee, & Podsakoff, 2003). To test if commonmethod variancehad contaminated the research findings, statistical procedures outlinedby Williams, Cote, and Buckley (1989) were undertaken.

Model 1, a baseline 3-factor measurement model [χ2(62) = 122.33,p b .001; CFI= .966, RMSEA= .054; SRMR= .048], was comparedwithModel 2, the baseline 3-factormeasurementmodel that included a com-mon latent factor [χ2(49)= 78.92, p b .001; CFI= .983, RMSEA= .043;SRMR = .035]. Model 2's fit with the data was a statistically significantimprovement overModel 1 (χ2 (13)= 43.4, p b .05)which denotes thepresence of commonmethod variance. The average amount of varianceaccounted for by three measures of interest was 46.7%. The averageamount of variance attributed to the common latent factor was 12.8%,a finding substantially less than the mean percentage (27%) reported

6 In addition to the chi-square statistic, Mueller and Hancock (2008) recommend thatresearchers select a single fit index from each class of indices: incremental (i.e., CFI;Bentler, 1990), parsimonious (i.e., RMSEA; Steiger, 1990), and absolute (i.e., SRMR;Bentler, 1995).

in Williams et al. (1989) empirical review. Further, the three bivariatecorrelations of interest retained their statistical significance despitethe addition of a common latent factor. These empirical results indicatethat the findings from this study can be attributed to factors other thanmethod variance.

6.3. Structural models

In addressing the two research hypotheses, four different structuralmodels were fit.,7,8 First, Karasek's (1979; Karasek & Theorell, 1990) ad-ditive model (Hypothesis 1) was tested by specifying direct paths fromperceived job control to perceived job strain, and fromperceived job de-mands to perceived job strain (Model 1). As shown in Table 3, Model 1produced an excellent fit with the data [χ2(73) = 145.25, p b .001;CFI = .952, RMSEA = .055; SRMR = .049]. The direct paths from per-ceived job control to perceived job strain (β = − .16, p b .05) and per-ceived job demands to perceived job strain (β = .52, p b .05) wereboth statistically significant in the predict direction (see Fig. 1). Thus,Model 1 offered empirical support for Hypothesis 1.

Next, a baseline full mediation model (Mode1 2) was tested(Hypothesis 2). Kenny, Kashy, and Bolger (1998) identified at leasttwo conditions that must be met when establishing a fully mediatedrelationship.9 First, the indirect path from the predictor variable to themediator variable, and the direct path from the mediator variable tothe criterion variable, must be statistically significant. Second, afterintroducing a direct path from the predictor variable to the criterionvariable into the structural model, the direct path from the mediatorvariable to the criterion variable, and the indirect path from the predic-tor variable to the mediator variable, should remain statistically signifi-cant. The direct path from the predictor variable to the criterionvariable, however, should be statistically nonsignificant.

Following this approach, Model 2 specified an indirect path fromperceived job control to self-report instrumental feedback, and a directpath from self-report instrumental feedback to perceived job strain. To

9 State-of-the-art thinking on testing for mediated effects no longer advocates estab-lishing a statistically significant correlation between the predictor and criterion variables.Kenny et al. (1998) point out that this relationship is implied when there is a statisticallysignificant predictor–mediator association and a statistically significant mediator-criterion association.

10 Bold beta coefficients are statistically significant.11 Bold beta coefficients are statistically significant.

Table 3AMOS fit measures for additive effects and mediational models.

Model χ2 df p-Value CFI RMSEA SRMR AIC

1 145.25 73 0.001 .952 .055 .049 237.252 232.36 130 0.001 .953 .049 .057 350.363 230.93 129 0.001 .954 .049 .056 350.934 277.02 130 0.001 .933 .058 .086 395.02

Note: Model 1 is an additive effects model.Model 2 is a full mediation model.Model 3 is partial mediation model.Model 4 is a non-nested alternative mediation model.

36 M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

properly test Karasek's (1979; Karasek & Theorell, 1990) additive hy-pothesis, an additional direct path from perceived job demands toperceived job strain was also included. Model 2 exhibited excellentfit with the data [χ2(130) = 232.36, p b .001; CFI = .953,RMSEA = .049; SRMR = .057] (see Table 3). Moreover, the indirectpath from perceived job control to self-report instrumental feedback(β = .32, p b .05) and the direct path from self-report instrumentalfeedback to perceived job strain (β=− .32, p b .05) were statistical-ly significant; as was the direct path from perceived job demands toperceived job strain (β = .55, p b .05).

Next, this study compared the baseline full mediation model(Model 2) with the partial mediation model (Model 3). Model 2and Model 3 were identical except for the addition of a direct pathfrom perceived job control to perceived job strain. Like Model 2,Model 3 also produced an excellent fit with the data [χ2(129) =230.93, p b .001; CFI = .954, RMSEA = .049; SRMR = .056]. Model3, however, was not a statistically significant improvement overModel 2 [χ2(2) = 1.43, n.s.] and its direct path from perceived jobcontrol to perceived job strain (β = − .08, n.s.) failed to achieve sta-tistical significance (see Fig. 2). The indirect path from perceived jobcontrol to self-report instrumental feedback (β = .32, p b .05), andthe two direct paths from self-report instrumental feedback to per-ceived job strain (β = − .30, p b .05) and perceived job demands toperceived job strain (β= .51, p b .05) maintained their statistical sig-nificance (see Fig. 2). These research findings suggest that Model 2 fitthe data better than Model 3 and, more importantly, are consistentwith full mediation (Hypothesis 2).

The robustness of this study's full mediation finding was exam-ined against a theoretically relevant non-nested alternative media-tion model. Social information processing theory (Salancik &Pfeffer, 1978) contends that the characteristics of an employee'sjob (e.g., perceptions of job control) are constructed from social in-formation (e.g., process and outcome feedback) derived from theirwider occupational environment (e.g., supervisors and co-workers). According to Salancik and Pfeffer, socially-constructedperceptions of one's job characteristics, as oppose to objective reali-ty, shape employees' affective experiences. For example, public childwelfare case managers who repeatedly hear their colleagues com-plain that unit caseloads are too high and difficult, are predicted toconstrue the demands of their job as overly taxing and strenuous, ir-respective of their assigned caseload's actual size or level ofcomplexity.

To test this theory-based non-nested alternative hypothesis, thecasual ordering for perceived job control and self-report instrumen-tal feedback present in Model 2 was reversed in Model 4. As shownin Table 3, Model 2 possessed a smaller AIC value (AIC = 350.36)than Model 4 (AIC = 395.02); thereby offering empirical evidencethat Model 2 fit the data best. Lastly, a 5000 sample bootstrappingtest was performed. The bootstrapping test, as with the prior SEManalyses, also yielded support for self-report instrumentalfeedback's role as a mediating variable (95% bias-corrected confi-dence interval = − .053, − .163). Overall, the patterning of empiri-cal findings for this particular sample of public child welfare casemanagers is consistent with full mediation.

7. Discussion

Public child welfare case managers practice in workplace environ-ments universally characterized as extremely stressful. Organizationaland occupational health researchers, across multiple social sciencedisciplines, assert that attenuating the strain perceptions of employeesin demanding employment settings, like child welfare, requires controlover one's job (Bakker & Demerouti, 2007; Karasek, 1979). In line withthis idea, the JD–C model's additive hypothesizes asserts that perceivedjobdemands andperceived job control concurrently impact perceptionsof job strain (Karasek & Theorell, 1990). Over three decades of cross-sectional, longitudinal, and experimental studies, however, have uncov-ered mixed research findings (Häusser et al., 2010; Van der Doef &Maes, 1999). This study sought to clarify inconsistent empirical resultsby testing self-report instrumental feedback's intervening role on thecontrol-strain relationship under perceptions of high job demands.

In exploring this potential mediated relationship, the JD–C model's(Karasek, 1979; Karasek & Theorell, 1990) basic theoretical rationalewas extended to incorporate the concept of indeterminate humanservice technologies (Hasenfeld, 1983, 2010) and prescriptions fromthe literature on dynamic complexity (Osman, 2010). Karasek's(1979) and Karasek & Theorell (1990) additive model proposes that,in tandem, perceived job demands and perceived job control have a sta-tistically significant, but opposite, main effect relationship with per-ceived job strain. Study data uncovered support for this hypothesis(Hypothesis 1). When self-report instrumental feedback was excludedfrom the SEM additive model analysis, the direct paths from perceivedjob demands to perceived job strain, and from perceived job control toperceived job strain, produced statistically significant main effects inthe expected direction (see Fig. 1).10

In contrast to Karasek (1979; Karasek & Theorell, 1990), the humanservices literature maintains that indeterminate technologies negateperceived job control's strain buffering attributes (Hasenfeld, 1983,2010). Further, the literature on complex and dynamic social environ-ments contends that quality performance outcomes necessitate percep-tions of control and ample goal-related information (Osman, 2010). Inaccordance with these ideas, data from this study indicated that self-report instrumental feedback fully mediated the relationship betweenperceived job control and perceived job strain (Hypothesis 2; seeFig. 2).11 When demands of the job are experienced as strenuous, per-ceived job control's strain inhibiting properties (i.e., facilitating thelearning of new knowledge, testing of new problem solving strategies,and expansion or refinement of existing job skills) appear whollyreliant on the ability of public child welfare case managers to secure in-strumental feedback from their larger occupational surroundings.Although sparse, comparable scientific findings can be found in the so-cial work literature. Pousette et al. (2003) reported that the relationshipbetweenperceived role ambiguity's (i.e., job-level uncertainty) and self-report job satisfactionwasmediated by the type of feedback (positive ornegative) human service workers received. Thus, as the first knownempirical work to reveal a fully mediated control–strain relationshipwithin Karasek's conceptual framework, this study's findings contributeto both the extant child welfare and JD–C literatures.

More importantly, study data challenge a core pillar of the JD–C(Karasek, 1979; Karasek & Theorell, 1990) model. Karasek argued thatthe “existence of a multiplicative interaction term is not the primaryissue” with respect to his model (1989, p. 143). He, instead, insistedthat the “primary ‘interaction’…in the [JD–Cmodel] is that two separatesets of outcomes are jointly predicted by two different combinations ofjob demands and decision latitude” (Karasek, 1989, p. 143). JD–C litera-ture reviews have uniformly reported mixed empirical evidence forKarasek's (1979; Karasek & Theorell, 1990) additive model on variousmeasures of physical and psychological strain (De Lange et al., 2003;

Fig. 1. Additive effects model.

37M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

Häusser et al., 2010; Van der Doef & Maes, 1999). This uneven body ofscientific evidence, has not prevented some occupational health psy-chologist from proclaiming that Karasek's (1979; Karasek & Theorell,1990) additive hypothesis is “no longer in doubt” and that “there is nofurther need for cross-sectional examination of main effects” (Häusseret al., 2010, p. 33). Empirical evidence from this study suggests thatwithin public child welfare's challenging occupational milieu, perceivedjob control may possess an indirect, as opposed to a main effect, rela-tionship with perceived job strain. Hence, when public child welfarecase managers judge their job as taxing, process and outcome(i.e., instrumental) feedback, rather than skill discretion or decisionlatitude, appear more central to advancing active learning.

This conclusion parallels findings from social cognitive theory(Bandura, 1986) and theories of self-regulated learning (Sitzmann &Ely, 2011). Bandura (1997), for example, notes that learning does not

Fig. 2. Partial med

arise in a vacuum but results from feedback information that emergeswhen individuals interact with their immediate social environment.Studyfindings also indirectly corroborate cross-sectional and experimen-tal JD–C research reporting contradictory individual-level outcomeswhen control was perceived as nominal. For instance, in a cross-sectional survey, Preston (2013b) found that when job demands wereperceived as high, public child welfare case managers who reportedlower amounts of job control and higher levels of instrumental feedback,experienced greater internal work motivation than their colleagues whoreported higher amounts of job control and lower levels of instrumentalfeedback.

Moreover, experimental studies by Jimmieson and Terry (1998a,b;1999) uncovered positive individual-level outcomes under conditionsof low task control and high procedural information. When proceduralinformation was received prior to engaging in a demanding task

iation model.

38 M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

activity, subjects assigned to the low task control condition reportedhigher levels of task satisfaction than subjects assigned to the hightask control condition (Jimmieson & Terry, 1998b). Based on this andsimilar experimental findings (Jimmieson & Terry, 1998a,b; 1999) con-cluded that job control's affective effects “may be contingent on…[hav-ing] access to information concerning various features of the workenvironment” (p. 366). In combination with existing social science the-ory and JD–C research, study data call into question the predictive valid-ity and practice utility of Karasek's (1979) and Karasek & Theorell(1990) seminal additive model for case managers practicing in publicchild welfare agencies.

8. Implications

Given the aforementioned, study data offer implications for both JD–C research and public sector child welfare practice. As previously noted,Hasenfeld (1983, 2010) argues that indeterminate technologies in-crease the amount of operational uncertainty public child welfare casemanagers experience at work. Osman (2010) states that sustaininghigh quality performance in complex and dynamic social settingsrequires personal control and goal-related information. These ideasare congruent with data from this study and, as such, may increasethe predictive validity of not only Karasek's (1979; Karasek & Theorell,1990) additive model, but also his JD–C interaction model. Support forthe latter argument can be found in a cross-sectional survey byHoukes, Janssen, de Jonge, and Nijhuis (2001). Using a sample ofhuman service (i.e., teachers) and non-human service (i.e., bank staff)employees, study data revealed a statistically significant and positivedemands–control interaction on internal workmotivation for bank em-ployees (i.e., predictable and reliable industrial-like technologies), and anull finding for teachers (i.e., indeterminate human servicetechnologies).

In regard to practice implications, study findings offer insights forimproving the practice utility of the JD–C (Karasek, 1979; Karasek &Theorell, 1990)model in public childwelfare agencies. Since instrumen-tal feedback facilitates perceived job control's strain-reducing proper-ties, public child welfare agencies should institutionalize formalchannels through which their case managers can obtain and learnfrom process and outcome feedback (Jimmieson & Terry, 1999;Preston, 2013a,b). Elected and appointed federal and state overseershave increasingly stipulated that public child welfare agencies collectperformance outcome data (Poertner, 2009). The federal government,for example, requires that public child welfare agencies store case-level data on all children under their legal care or supervision (Pecoraet al., 2012). Disaggregating and analyzing this data at the office, district,region, county, and state level can strengthen public child welfare casemanagers' performance expectations by providing them with outcomefeedback on which types of service interventions or menu of interven-tions will (or have) produced the best client outcomes (Antle,Christensen, Van Zyl, & Barbee, 2012).

Although outcome data are useful for learning whether or not man-dated performance standards can be (or have been) attained, this typeof feedback information does not enlighten public child welfare casemanagers with meaningful procedural knowledge on areas of practicedeficiency or proficiency (Courtney, Needell, & Wulczyn, 2004).Without this type of process feedback, strengthening perceptions ofjob competence is extremely difficult (Earley et al., 1990). Process(and outcome) feedback is most useful when it is provided in a timelymanner from multiple, creditable, and reliable sources (Ilgen, Fisher, &Taylor, 1979). Evidence-based practices that can supply public childwelfare case managers with this type of feedback information includesupervisory case staffings (Barak, Dnika, Pyun, & Bin, 2009; Moyle,1998); consultations with outside experts; and/or case conferencesthat include relevant contracted social and human service providers,oversight organizations, and family members (Crea & Berzin, 2009;Pecora et al., 2012). Finally, the more specific the process feedback

received, the easier it is for public child welfare case managers to deci-pher which job skills, problem solving strategies, and knowledge gainswere (in)effective andwhy; what contextual factors led to concomitantperformance errors; and what types of corrective action are needed(Goodman, Wood, & Hendrickx, 2004)

9. Limitations and future research

Limitations include this study's data collection method and theomission of a core JD–C construct. Cross-sectional survey data preventsone from making definitive statements on the directionality of thecausal relationship between the main variables of interest (Hair et al.,2006), while the ability to more fully test Karasek's (1979; Karasek &Theorell, 1990) additive model was impeded by the omission of an ac-tive learning variable. Finally, the racial and gender composition of pub-lic child welfare case managers, and geographic homogeneity of theirhost agencies, cautions against positing national and internationalclaims with respect to this study findings. Future research should seekto replicate these unique empirical results on other employee-relatedoutcomes of relevance to the field of child welfare. Job satisfaction, forexample, has been identified as problematic in public child welfareagencies (Barth et al., 2008). Because perceived control's attitudinaleffects are derived from environmental information (Skinner, 1995),self-report instrumental feedback should also fully mediate perceivedjob control's association with perceived job satisfaction.

10. Conclusion

The extant JD–C literature has consistently produced mixed empiri-cal results with respect to Karasek's additive hypothesis (De Lange et al.,2003; Häusser et al., 2010; Taris & Kompier, 2005a,b; Van der Doef &Maes, 1999). This study addressed this issue by testing self-reportinstrumental feedback'smediating role on the association between per-ceived job control and perceived job strain. In line with Karasek's addi-tive model, perceived job demands produced a statistically significantpositive main effect and perceived job control produced a statisticallysignificant negative main effect on perceived job strain. However,when self-report instrumental feedbackwas included as an interveningvariable, the direct path from perceived job control and perceived jobstrain was no longer statistically significant. Consequently, researchfindings advance both the childwelfare and JD–C literatures in two sub-stantive ways. Data from this study call into question the predictive va-lidity of Karasek's (1979; Karasek & Theorell, 1990) additive model bydemonstrating that self-report instrumental feedback, as opposed toperceived control over one's job, may function as a more proximalcausal mechanism through which active learning mitigates the strainperceptions of public child welfare case managers. Further, study dataindirectly suggest that integrating the concept of indeterminate tech-nologies and evidence from the literature on complex and dynamic en-vironments into Karasek's (1979; 1998; Karasek & Theorell, 1990)theoretical logic may strengthen the additive model's practice utilityin public child welfare agencies.

References

Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52(3), 317–332.Aldrich, H. (2008). Organizations and environments. Stanford University Press.Anderson, J., & Gerbing, D. (1992). Assumptions and comparative strengths of the two-

step approach comment on Fornell and Yi. Sociological Methods & Research, 20(3),321–333.

Antle, B., Christensen, D., Van Zyl, M., & Barbee, A. (2012). The impact of the SolutionBased Casework (SBC) practice model on federal outcomes in public child welfare.Child Abuse & Neglect, 36(4), 342–353.

Austin, M. (2002). Human services management: Organizational leadership in social workpractice. New York: Columbia University Press.

Bandura, A. (1986). Social Foundations of Thought and Action. Englewood Cliffs, NJ:Prentice Hall.

Bandura, A. (1997). Self-Efficacy: The Exercise of Control. Macmillan.

39M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

Barak, M., Dnika, T., Pyun, H., & Bin, X. (2009). The impact of supervision on worker out-comes: A meta‐analysis. Social Service Review, 83(1), 3–32.

Baron, R., & Kenny, D. (1986). The moderator–mediator variable distinction in socialpsychological research: Conceptual, strategic, and statistical considerations. Journalof Personality and Social Psychology, 51(6), 1173–1182.

Barth, R., Lloyd, E., Christ, S., Chapman, M., & Dickinson, N. (2008). Child welfareworker characteristics and job satisfaction: A national study. Social Work,53(3), 199–209.

Becker, M. (2004). Organizational routines: A review of the literature. Industrial andCorporate Change, 13(4), 643–678.

Becker, M., & Knudsen, T. (2005). The role of routines in reducing pervasive uncertainty.Journal of Business Research, 58(6), 746–757.

Bentler, P. (1990). Comparative fit indexes in structural models. Psychological Bulletin,107(2), 238–246.

Bentler, P. (1995). EQS structural equations program manual. Encino, CA: MultivariateSoftware.

Bergman, P., Ahlberg, G., Johansson, G., Stoetzer, U., Åborg, C., Hallsten, L., et al. (2012). Dojob demands and job control affect problem-solving? Work: A Journal of Prevention,Assessment and Rehabilitation, 42(2), 195–203.

Caplan, R., Cobb, S., French, J., Van Harrison, R., & Pinneau, S. (1975). Demands and workerhealth: Main effects and organizational differences. Washington, DC: US GovernmentPrinting Office.

Courtney, M., Needell, B., & Wulczyn, F. (2004). Unintended consequences of the push foraccountability: The case of national child welfare performance standards. Childrenand Youth Services Review, 26(12), 1141–1154.

Crea, T., & Berzin, S. (2009). Family involvement in child welfare decision-making: Strat-egies and research on inclusive practices. Journal of Public Child Welfare, 3(3),305–327.

Dane, B. (2000). Child welfare workers: An innovative approach for interacting withsecondary trauma. Journal of Social Work Education, 36(1), 27–38.

Daniels, K., Beesley, N., Wimalasiri, V., & Cheyne, A. (2013). Problem Solving and Well-Being Exploring the Instrumental Role of Job Control and Social Support. Journal ofManagement, 39(4), 1016–1043.

Daniels, K., Boocock, G., Glover, J., Hartley, R., & Holland, J. (2009). An experience samplingstudy of learning, affect, and the demands control support model. Journal of AppliedPsychology, 94(4), 1003–1017.

Darlington, Y., Feeney, J., & Rixon, K. (2004). Complexity, conflict and uncertainty: Issuesin collaboration between child protection and mental health services. Children andYouth Services Review, 26(12), 1175–1192.

De Jonge, J., & Kompier, M. (1997). A critical examination of the demand-control-supportmodel from a work psychological perspective. International Journal of StressManagement, 4(4), 235–258.

De Jonge, J., Spoor, E., Sonnentag, S., Dormann, C., & van den Tooren, M. (2012). “Take abreak?!” Off-job recovery, job demands, and job resources as predictors of health,active learning, and creativity. European Journal of Work and OrganizationalPsychology, 21(3), 321–348.

De Lange, A., Taris, T., Kompier, M., Houtman, I., & Bongers, P. (2003). The very best of themillennium: Longitudinal research and the demand-control-(support) model. Journalof Occupational Health Psychology, 8(4), 282–305.

DePanfilis, D., & Zlotnik, J. (2008). Retention of front-line staff in child welfare: A system-atic review of research. Children and Youth Services Review, 30(9), 995–1008.

Dillman, D. (2007). Mail and internet surveys: The tailored design,—2007 update. Hoboken:John Wiley.

Earley, C., Northcraft, G., Lee, C., & Lituchy, T. (1990). Impact of process and outcome feed-back on the relation of goal setting to task performance. Academy of ManagementJournal, 33(1), 87–105.

Frese, M., Brodbeck, F., Heinbokel, T., Mooser, C., Schleiffenbaum, E., & Thiemann, P.(1991). Errors in training computer skills: On the positive function of errors.Human–Computer, Interaction, 6, 77–93.

Frese, M., & Stewart, J. (1984). Skill learning as a concept in life-span developmentalpsychology: An action theoretic analysis. Human Development, 27(3-4),145–162.

Galbraith, J. (1973). Designing Complex Organizations. Addison-Wesley Longman Publish-ing Co., Inc.

Gambrill, E. (2008). Decision making in child welfare: Constraints and potentials. In D.Lindsey, & A. Shlonsky (Eds.), Child welfare research: Advances for practice and policy(pp. 175–193). New York: Oxford University Press.

Goodman, J., Wood, R., & Hendrickx, M. (2004). Feedback specificity, exploration, andlearning. Journal of Applied Psychology, 89(2), 248–262.

Greenglass, E., Fiksenbaum, L., & Burke, R. (1996). Components of social support, bufferingeffects and burnout: Implications for psychological functioning. Anxiety, Stress, andCoping, 9(3), 185–197.

Hackman, J., & Oldham, G. (1980). Work redesign. Reading, MA: Addison-Wesley.Hair, J., Black, W., Babin, B., Anderson, R., & Tatham, R. (2006). Multivariate data analysis

(6th ed.). Upper Saddle River, NJ: Prentice Hall.Hasenfeld, Y. (1983). Human Service Organizations. Englewood Cliffs, NJ: Prentice-Hall.Hasenfeld, Y. (2010). The attributes of human service organizations. In Y. Hasenfeld

(Ed.), Human services as complex organizations (pp. 9–32). Thousand Oaks, CA:Sage.

Häusser, J., Mojzisch, A., Niesel, M., & Schulz-Hardt, S. (2010). Ten years on: A review ofrecent research on the job demand-control (-support) model and psychologicalwell-being. Work and Stress, 24(1), 1–35.

Hayes, A. (2009). Beyond Baron and Kenny: Statistical mediation analysis in the newmil-lennium. Communication Monographs, 76(4), 408–420.

Houkes, I., Janssen, P., de Jonge, J., & Nijhuis, F. (2001). Specific relationships betweenwork characteristics and intrinsic work motivation, burnout and turnover intention:

A multi-sample analysis. European Journal of Work and Organizational Psychology,10(1), 1–23.

Ilgen, D., Fisher, C., & Taylor, S. (1979). Consequences of individual feedback on behaviorin organizations. Journal of Applied Psychology, 64(4), 349–371.

Jackson, S. (1989). Does job control control stress? In S.L. Sauter, J.J. HurrellJr., & C.L.Cooper's (Eds.), Job control and worker health (pp. 25–53). Chichester: Wiley.

Jimmieson, N., & Terry, D. (1998a). An experimental study of the effects of work stress,work control, and task information on adjustment. Applied Psychology, 47(3),343–369.

Jimmieson, N., & Terry, D. (1998b). Job-related information as a buffer of work stress.Proceedings of the 1st international work psychology conference, 1–3 July 1998. Instituteof work psychology, Sheffield, UK: University of Sheffield. University of Sheffield,Institute of work psychology.

Jimmieson, N., & Terry, D. (1999). Themoderating role of task characteristics in determin-ing responses to a stressful work simulation. Journal of Organizational Behavior, 20(5),709–736.

Kain, J., & Jex, S. (2010). Karasek's (1979) job demands–control model: A summary of cur-rent issues and recommendations for future research.New developments in theoreticaland conceptual approaches to job stress, 8. (pp. 237–268).

Karasek, R. (1979). Job demands, job decision latitude, andmental strain: Implications forjob redesign. Administrative Science Quarterly, 24(2), 285–308.

Karasek, R. (1989). Control in the workplace and its health related aspects. In S.L.Sauter, J.J. HarrellJr., & C.L. Cooper (Eds.), Job control and worker health. Chiches-ter: Wiley.

Karasek, R. (1998). Demand/control model: A social, emotional, and physiologicalapproach to stress risk and active behaviour development. Encyclopedia ofOccupational Health and Safety, 2. (pp. 34.6–34.14).

Karasek, R., & Theorell, T. (1990). Healthy Work. New York: Basic Book.Kenny, D., Kashy, D., & Bolger, N. (1998). Data analysis in social psychology. The Handbook

of Social Psychology, 1(4), 233–265.Kim, H. (2011). Job conditions, unmet expectations, and burnout in public child welfare

workers: How different from other social workers? Children and Youth ServicesReview, 33(2), 358–367.

Kim, H., & Stoner, M. (2008). Burnout and turnover intention among social workers:Effects of role stress, job autonomy and social support. Administration in SocialWork, 32(3), 5–25.

Kluger, A., & DeNisi, A. (1996). The effects of feedback interventions on performance: Ahistorical review, a meta-analysis, and a preliminary feedback intervention theory.Psychological Bulletin, 119(2), 254–284.

Littell, J. (2005). Lessons from a systematic review of effects of multisystemic therapy.Children and Youth Services Review, 27(4), 445–463.

Littell, J.H. (2008). Evidence-based or biased? The quality of published reviewsof evidence-based practices. Children and Youth Services Review, 30(11),1299–1317.

Luchman, J., & González-Morales, M. (2013). Demands, control, and support: A meta-analytic review of work characteristics interrelationships. Journal of OccupationalHealth Psychology, 18(1), 37.

March, J., & Simon, H. (1958). Organizations. New York: Wiley.Marshall, N., Barnett, R., & Sayer, A. (1997). The changing workforce, job stress, and psy-

chological distress. Journal of Occupational Health Psychology, 2(2), 99–107.Moyle, P. (1998). Longitudinal influences of managerial support on employee well-being.

Work and Stress, 12(1), 29–49.Mueller, R., & Hancock, G. (2008). Best practices in structural equation modeling. In J.W.

Osborne (Ed.), Best practices in quantitative methods (pp. 488–508). Thousand Oaks,CA: Sage.

National Association of Social Workers (2004). “If you're right for the job, it's the bestjob in the world”: The National Association of Social Worker's Child WelfareSpecialty Practice members describe their experiences in child welfare. Washington,DC: Author (Retrieved from (http://www.naswdc.org/practice/children/NASWChildWelfareRpt062004.pdf).

Ohly, S., Sonnentag, S., & Pluntke, F. (2006). Routinization, work characteristics and theirrelationships with creative and proactive behaviors. Journal of OrganizationalBehavior, 27(3), 257–279.

Osman, M. (2010). Controlling uncertainty: A review of human behavior in complexdynamic environments. Psychological Bulletin, 136(1), 65–86.

Pecora, P., Whittaker, J., Maluccio, A., & Barth, R. (2012). The child welfare challenge: Policy,Practice, and Research. Aldine Transaction.

Podsakoff, P., MacKenzie, S., Lee, J., & Podsakoff, N. (2003). Common method biases inbehavioral research: A critical review of the literature and recommended remedies.Journal of Applied Psychology, 88(5), 879–903.

Poertner, J. (2009). Managing for service outcomes. In R. Patti's (Ed.), The Handbookof Social Welfare Management (pp. 267–281). Thousand Oaks, CA: SagePublications.

Pousette, A., Jacobsson, C., Thylefors, I., & Hwang, C. (2003). The role of feedback inSwedish human service organizations. Community, Work & Family, 6(3),245–268.

Preacher, K., Rucker, D., & Hayes, A. (2007). Addressing moderated mediation hypothe-ses: Theory, methods, and prescriptions. Multivariate Behavioral Research, 42(1),185–227.

Preston, M. S. (2007). Karasek's job demands–control model: A multi-method studyexamining the predictive validity of instrumental feedback as a second-order moderatorvariable. Unpublished doctoral dissertation. University at Albany, State University ofNew York, Albany, NY.

Preston, M.S. (2013a). Motivating child welfare case managers: An application andextension of feedback information theory. Children and Youth Services Review,35(4), 734–741.

40 M.S. Preston / Children and Youth Services Review 54 (2015) 30–40

Preston, M.S. (2013b). Advancing case manager motivation in child welfare: Job control'scurvilinear relationship and instrumental feedback's moderating influence. Childrenand Youth Services Review, 35(12), 2003–2012.

Rafferty, Y., Friend, R., & Landsbergis, P. (2001). The association between job skilldiscretion, decision authority and burnout. Work & Stress, 15(1), 73–85.

Raghunathan, R., & Pham, M. (1999). All negative moods are not equal: Motivational in-fluences of anxiety and sadness on decision making. Organizational Behavior andHuman Decision Processes, 79(1), 56–77.

Rapoport, R. (1960). Community as Doctor: New perspectives on a therapeutic community.London: Tavistock Publication.

Rzepnicki, T., & Johnson, P. (2005). Examining decision errors in child protection: A newapplication of root cause analysis. Children and Youth Services Review, 27(4), 393–407.

Salancik, G., & Pfeffer, J. (1978). A social information processing approach to job attitudesand task design. Administrative Science Quarterly, 23(2), 224–253.

Sandfort, J. (2010). Human service organizational technology: Improving understandingand advancing research. In Y. Hasenfeld (Ed.), Human services as complex organiza-tions (pp. 269–290) (2nd ed.). Newbury Park, CA: Sage Publications.

Seligman, M. (1975). Helplessness: On depression, development, and death. WH Freeman/Times Books/Henry Holt & Co.

Selye, H. (1950). Stress. Montreal: Acta.Sitzmann, T., & Ely, K. (2011). A meta-analysis of self-regulated learning in work-related

training and educational attainment: What we know and where we need to go.Psychological Bulletin, 137(3), 421–442.

Skinner, E. (1995). Perceived Control, Motivation, and Coping. Sage Publications, Inc.Smith, B. (2010). Service technologies and the condition of work in child welfare. In Y.

Hasenfeld (Ed.), Human services as complex organizations (pp. 253–268) (2nd ed.).Newbury Park, CA: Sage Publications.

Sobel, M. (1982). Asymptotic confidence intervals for indirect effects in structuralequations models. In S. Leinhart (Ed.), Sociological methodology (pp. 290–312). SanFrancisco: Jossey-Bass.

Söderfeldt, B., Söderfeldt, M., Muntaner, C., O'Campo, P., Warg, L., & Ohlson, C. (1996). Psy-chosocial work environment in human service organizations: A conceptual analysisand development of the demand-control model. Social Science and Medicine, 42(9),1217–1226.

Sosin, M. (2010). Discretion in human service organizations: Traditional and institutionalperspectives. In Y. Hasenfeld (Ed.), Human services as complex organizations(pp. 381–403) (2nd ed.). Los Angeles, CA: Sage.

Steiger, J. (1990). Structural model evaluation and modification: An interval estimationapproach. Multivariate Behavioral Research, 25(2), 173–180.

Taris, T., & Kompier, M. (2005a). Job demands, job control, strain and learning behavior:Review and research agenda. In A.G. Antoniou, & C.L. Cooper (Eds.), Research Compan-ion to Organizational Health Psychology (pp. 132–150). London: Edward Elgar Press.

Taris, T., & Kompier, M. (2005b). Job characteristics and learning behavior: Review and 34psychological mechanisms. In P. Perrewé, & D. Ganster's (Eds.), Research in Occupa-tional Stress and Well Being: Exploring interpersonal dynamics. Vol. 4. (pp. 127–166).Amsterdam: JAI Press.

Terry, D., & Jimmieson, N. (1999). Work control and employee well-being: A decadereview. In C. Cooper, & I. Robertson's (Eds.), International Review of Industrial andOrganizational Psychology, Vol. 14. (pp. 95–148). Chichester, UK: Wiley.

Van der Doef, M., & Maes, S. (1999). The job demand-control (-support) model andpsychological well-being: A review of 20 years of empirical research. Work andStress, 13(2), 87–114.

Van Katwyk, P., Fox, S., Spector, P., & Kelloway, K. (2000). Using the Job-Related AffectiveWell-Being Scale (JAWS) to investigate affective responses to work stressors. Journalof Occupational Health Psychology, 5(2), 219–230.

Wall, T., Jackson, P., Mullarkey, S., & Parker, S. (1996). The demands–control model of jobstrain: A more specific test. Journal of Occupational and Organizational Psychology,69(2), 153–166.

Warr, P. (1990). The measurement of well‐being and other aspects of mental health.Journal of Occupational Psychology, 63(3), 193–210.

Warr, P., & Downing, J. (2000). Learning strategies, learning anxiety and knowledgeacquisition. British Journal of Psychology, 91(3), 311–333.

Weston, R., & Gore, P. (2006). A brief guide to structural equation modeling. TheCounseling Psychologist, 34(5), 719–751.

White, H. (1980). A heteroskedastic consistent covariance matrix and a direct test forheteroskedasticity. Econometria, 48(4), 817–838.

Wielenga-Meijer, E., Taris, T., Kompier, M., & Wigboldus, D. (2010). From task character-istics to learning: A systematic review. Scandinavian Journal of Psychology, 51(5),363–375.

Williams, L., Cote, J., & Buckley, M. (1989). Lack of method variance in self-reported affectand perceptions at work: Reality or artifact? Journal of Applied Psychology, 74(3),462–468.

Williams, L., Vandenberg, R., & Edwards, J. (2009). 12 Structural equation modeling inmanagement research: A Guide for Improved Analysis. The Academy of ManagementAnnals, 3(1), 543–604.