Explaining Popular Trust in the Departmentof Homeland Security

Scott E. RobinsonXinsheng Liu

James StoutenboroughArnold Vedlitz

Bush School of Government and Public ServiceInstitute for Science, Technology, and Public Policy

Texas A&M University

DRAFT: PLEASE DO NOT QUOTEWITHOUT THE AUTHOR’S PERMISSION

October 11, 2011

1

Abstract

Research reveals that levels of reported trust in government are ata relatively low level – among the lowest in the period studied. Atthe same time, reported approval for specific administrative agenciesvaries widely with some agencies receiving little support and others agreat deal. This raises an important question; what factors drive trustin specific agencies? This article investigates the question in relation tothe Department of Homeland Security (DHS). We find that reportedassessments of DHS are driven by political attitudes, policy salience,religiosity, and demographic characteristics – even when controllingfor trust in government in general.

2

1 Introduction

In a democratic society, it is essential that government maintain trust andapproval from the public (e.g. Dahl 1971, Easton 1965, Putnam 2001). This isthe case even for components of the government, like administrative agenciesand the courts, that may not have direct electoral accountability. Questionsof government approval, then, become important to research in public admin-istration generally. This has led to a great deal of research into the factorsrelated to trust in government, including a large literature on the impact ofe-government on various evaluations of government.

While most of this literature has focused on factors related to govern-ment in general or specific branches of government, there is reason to believethat there are many differences between the evaluations of different specificadministrative agencies. Polls conducted by the Pew Foundation have peri-odically asked respondents for assessments of specific administrative agencies(in addition to more traditional questions related to government in general,the president, Congress, etc.) including a recent iteration(Pew ResearchCenter 2010). This survey revealed considerable variation in assessments ofspecific agencies. The IRS routinely comes in with low ratings of generalapproval, though with some improvements in the most recent survey. TheDepartment of Education fell to become the lowest rated agency in this mostrecent survey. In all, this makes clear that respondents differentiated theirassessments of specific agencies. Given this variability, we are interested inagency-specific models of trust and approval.

This article builds a model of approval for the Department of HomelandSecurity (DHS). In addition to the factors commonly accepted as importantto trust in government in general, we propose some additional demographicvariables that may explain trust in DHS and assess the impact of the salienceof terrorism on trust in DHS. The findings provide some insight into theextent to which respondents differentiate their assessments of agencies andthe factors that play a substantively important, but historically neglected,role in explaining these trust assessments.

2 Trust of Government

In recent decades there has been a overall trend of decreasing public trust ingovernment and political institutions in the United States. This decline of

3

trust has also been found in many advanced democracies (Dalton and Wat-tenberg 2000; Dalton 2004), in some newly emerged democracies (Catterberg& Moreno 2006), and in various developing countries (Cheema 2005). In theUnited States, various studies of public opinion demonstrate that the long-term falling trend in the level of Americans’ trust in government began inthe mid- to late 1960s (Bok 1997, Jennings 1998, Alford 2001, Hetherington2005), and this downward pattern seems to continue despite short-lived resur-gences in the 1980s and late 1990s as well as following the terrorist attackson September 11, 2001 (Jones 2006).

Many scholars in political science, sociology, and public opinion stud-ies have examined different sources of public trust (e.g. Blendon, Benson,Morin, Altman, Brodie, Brossard & James 1997, Caldeira & Gibson 1992,Chanley 2002, Chanley, Rudolph & Rahn 2000, Craig 1993, Erber & Lau1990, Feldman 1983, Garment 1991, Gibson, Caldeira & Spence 2003, Miller& Borrelli 1991, Orren 1997, Stoutenborough & Haider-Markel 2008, Williams1985), various consequences of distrust (Hetherington 1998, Hetherington1998, Hetherington 2005, Hetherington 1999) and major benefits of highcitizens’ trust for democratic society (e.g. Dahl 1971, Easton 1965, Easton1975, Keele 2007, Putnam 1993, Putnam 2001), public policy and publicadministration (Hetherington 2005), and even compliance with laws (Levi1997, Scholz & Lubell 1998, Scholz & Pinney 1995, Tyler 1990, Tyler &Degoey 1995). These studies have greatly contributed to our understandingof citizens’ trust/distrust in government. However, there are two key gaps inextant research.

First, there has been little attention to trust in specific administrativeagencies. Regardless of the precise wording of the questions asked in variouspublic surveys, most studies have focused on citizens’ holistic/general evalu-ation of the entire political system or specific governmental branches. Thereis no doubt that macro-level holistic models of public trust are useful, partic-ularly in examining the overall trend of public evaluation of a political regime(Bok 1997, Hetherington 2005), or in comparing citizens’ assessments of theoverall performance across different governmental levels (e.g., federal, state,and local governments; see Jennings 1998) or across different governmentalbranches (Jones 2006). However, the findings from this type of research pro-vide little insights about the relationship between public citizens’ trust andspecific administrative agencies. Given the observed variation in approval forspecific agencies, this is an important limitation of existing research.

Conceptually, citizens’ generalized trust in government or governmental

4

branches and citizens’ trust in specific administrative agency are distinct no-tions of trust. A generalized low level of public trust in the entirety of thepolitical system or federal government does not necessarily mean low trustor low confidence in all governmental branches. For instance, public opinionpolls have consistently found that Americans express greater trust in the ju-dicial branch than in the executive and legislative branches (e.g. Jones 2006,Lipset & Schneider 1983, Stoutenborough & Haider-Markel 2008). Like-wise, a low level of the overall trust in the executive branch does not nec-essarily mean low public trust in all administrative agencies (Pew ResearchCenter 1998, Pew Research Center 2010). In reality, citizens understand thatthere are multiple components of the federal government, and these compo-nents perform different functions and touch their lives in different ways. De-pending on one’s specific situation and personal characteristics as well as theparticular function, conduct and performance of the respective components,individual citizens often find themselves with varying assessments of differentcomponents of the federal government.

In a recent study on e-government-citizen trust relationship, Morgenson,VanAmburg and Mithas (2011) highlighted the importance of differentiatingbetween system- or branch-based general trust and citizens’ particular trustin specific agencies. They measured specific trust in a total of 55 distinctfederal agencies or departments. However, as all the data of the specific trustin these agencies / departments were used in a pooled fashion, their study didnot provide information on what factors influence individual citizens’ trust inspecific agencies. In the end, Keele is correct to note that the general trustin government literature’s ”findings are easily summarized by saying thattrust is a reflection of government performance” (2007, 242). Unfortunately,it is unclear to what extent this will help us to understand trust in specificagencies.

The second gap in extant literature is that most studies focus on orga-nizational / institutional determinants of public trust in government. Thisapproach emphasizes that the performance and conduct of the institution/ organization are the fundamental determinants of citizens’ trust/distrust(Keele 2007). For instance, Jennings (1998) found that citizens’ trust is posi-tively associated with three institutional/organizational characteristics: gov-ernment performance, citizen-government linkage (e.g., accessibility, trans-parency, etc.), and integrity of governmental officials. Other studies showsimilar findings that citizens tend to have higher trust/confidence when thegovernment manages the economy well (e.g. Chanley, Rudolph & Rahn 2000,

5

Citrin & Green 1986, Citrin & Luks 2001, Feldman 1983, Hetherington 1998,Keele 2007, Lawrence 1997, Miller & Borrelli 1991, Miller 1991, Williams1985), controls crime (e.g. Chanley, Rudolph & Rahn 2000, Mansbridge 1997,Pew Research Center 1998), avoids scandal and displays high levels of hon-esty and integrity (e.g. Black & Black 1994, Blendon et al. 1997, Chanley,Rudolph & Rahn 2000, Garment 1991, Keele 2007, Lipset & Schneider 1983,Orren 1997). The recent studies of e-government demonstrate that strongercitizen-government linkages (accessibility, transparency, interactivity) pro-vided by various forms and aspects of e-government also has positive effectson public trust and confidence (Welch, Hinnant & Moon 2005, Tolbert &Mossberger 2006, Morgeson, VanAmburg & Mithas 2011).

While extant research emphasizes the causes of trust from an institutionalperformance perspective, most studies tend to overlook the potential individ-ual bases of citizens’ trust. One critical/central question starting to get moreattention in the extensive literature of trust is whether individual trust levelsand variations in these levels are linked to individual characteristics. Thisresearch is starting to show that citizens’ ideological orientation (Rudolph &Evans 2005, Rudolph 2009) and party identifications (Gershtenson, Ladewig& Plane 2006, Fiorina & Abrams 2008, Bafumi & Shapiro 2009) affect theirtrust in government.

Other personal dimensions that researchers see growing in importanceinclude the religiosity of citizens. Recent research is finding that fundamen-talist religiosity is associated with many political attitudes and policy choices(Berkman & Plutzer 2009, Bolzendahl & Brooks 2005, Brooks 2002, Camp-bell & Monson 2008, Houston, Freeman & Feldman 2008, Sherkat 2011,Stokes & Ellison 2010). Additionally, citizens’ attention to national securityissues (Baumgartner, Francia & Morris 2008, Froese & Mencken 2009) andcitizens with stronger connectedness to other social elements (Putnam 2001)are evidencing higher trust than those who are not as concerned or who arenot closely connected to other people.

In what follows, we expand the discussion about the roots of public trustin government by conduct an agency-specific analysis emphasizing variousindividual-level sources of trust. More specifically, our objective is to examineindividual bases of public trust in one specific administrative agency - theDepartment of Homeland Security (DHS).

6

3 Theoretical Model of Trust of Administra-

tive Agencies

Given the diverse literature focused on trust in government, it is difficult tobuild a single model to unite all of the disparate explanatory components. Wewill start building a model by looking at the types of variables that previousstudies have included.

Previous models of trust in government have focused on generalized trustin government. The typical survey questions ask how often respondents trust“government in Washington DC” (e.g. Tolbert & Mossberger 2006) or apooled series of assessments of a wide variety of agencies (e.g. Morgeson,VanAmburg & Mithas 2011). Based on the general nature of these pools andquestion wording strategies, it is not unexpected that explanatory variablesgenerally consist of demographic variables. These assessments are fixed foreach individual and may affect their general disposition towards trusting gov-ernment agencies. Previous research has illustrated the impact of gender (e.g.Brewer & Sigelman 2002, Cook & Gronke 2005, Hetherington 1998, Keele2005), age (e.g. Hetherington 1998, Keele 2005, Welch, Hinnant & Moon2005), race (e.g. Brewer & Sigelman 2002, Hetherington 1998, Keele 2005,Tolbert & Mossberger 2006), education (e.g. Brewer & Sigelman 2002, Cook& Gronke 2005, Hetherington 1998), and income (e.g. Hetherington 1998).General demographic variables are included in just about every study –though the exact demographic components vary from model to model. Thisbasic components suggests that trust is a product of demographic variables(Λ)

Trust = f(gender, age, race, income, # of children, education, religiosity) =f(Λ)

Interestingly, few studies in the public administration literature have in-cluded controls for political ideology or partisan identification. These vari-ables have been of central importance to studies of generalized trust in gov-ernment in studies within political science (e.g. Brewer & Sigelman 2002,Keele 2005). It seems obvious that one’s trust in government (in a generalsense or as specifically applied to a single organization) would be affected byone’s political dispositions. This calls for a second component for the trustmodel: political attitudes (Γ).

Trust = f(Λ + Political Ideology, Party Identification) = f(Λ + Γ)

7

Given that our interests are in the specific evaluations of DHS, we willalso investigate whether respondents’ attention to terrorism as a public policyproblem affects their expected levels of trust in DHS (β). It may be thatpeople who report systematically higher levels of attention to terrorism havedifferent assessments of DHS. The attentive audiences are likely to knowmore about DHS activities and assessments of their activities. With thisadditional knowledge and the predispositions that made these audiences paymore attention to issues related to terrorism in the first place, one mightexpect their trust assessments to be different than respondents who reportedlower levels of attention to terrorism.

Trust = f(Λ + Γ + Attention to Terrorism)= f(Λ + Γ + β)

3.1 Key Hypotheses

We will focus on a few key hypotheses. While controlling for key demographicvariables, we will focus on the role of religiosity, political ideology, partyidentification and attention to terrorism. We have elected to spotlight thesevariables because they have not been looked at together in explaining agency-specific trust issues. This leads to the following hypotheses.

Hypothesis 1: The greater the individual religiosity of a respon-dent, the greater the levels of trust in DHS.

Hypothesis 2: The greater the conservative ideology of a respon-dent, the greater the level of trust in DHS.

Hypothesis 3: Respondents reporting Republican party identifi-cation of partisan independence will report greater levels of trustin DHS.

Hypothesis 4: The greater the respondent’s attention to terror-ism, the greater the level of trust in DHS.

The role of generalized trust in government is a more complex additionto some of the models. When asked to assess DHS, it is possible that generalpolitical trust drives respondent assessments. It may be that factors relatedto general political trust dominate the model of trust in DHS. To eliminatethis possibility, we include a measure of general political trust to control for

8

its direct effect on the specific assessments of DHS. We expect the effects ofthe previously discussed components (religiosity, general political attitudes,and policy salience) to remain even after controlling for the direct effect ofgeneralized political trust. This leads to our final hypothesis.

Hypothesis 5: The greater the respondent’s general political trust,the greater the level of trust in DHS.

4 Data and Methods

Testing these hypotheses requires data from respondents across a broad rangeof demographic characteristics and using complex statistical analysis to in-vestigate nuanced patterns within the data. This section details our survey-based data strategy, measurement approach, and data analysis strategy.

4.1 Survey Sample and Protocol

The analysis uses the results of a national public opinion survey focusedon issues related to terrorism and homeland security. The survey was con-ducted in August 2009 by the Public Policy Research Institute at Texas A&MUniversity. The telephone survey averaged about 35 minutes each and 924interviews were completed.1

All of the results reported below were calculated in STATA version 11(Statacorp 2009).

1Following American Association for Public Opinion Research conventions and algo-rithms, the response rate was 5.4%, the cooperation rate was 16.8% and the completionrate was 78.4%. The declining trend of response rates in recent polls has been carefullyexamined by survey scientists. Contrary to the conventional presumption that a lowerresponse rate leads to poorer survey quality, recent empirical studies indicate there are lit-tle statistical differences between survey results with high response rate and low responserate. In a comprehensive study using data drawn from exit polls, (Merkle & Edelman 2002)found no relationship between response rate and survey accuracy. (Keeter, Kennedy, Di-mock, Best & Craighill 2006) found that results from survey with lower responses ratewere generally statistically indistinguishable from those with much higher response rates.In another study comparing 81 national surveys with response rates varying from 5% to54%, (Holbrook, Krosnick & Pfent 2007) found that RDD telephone surveys with lowresponse rates “do not notably reduce the quality of survey demographic estimates.”

9

Figure 1: The Components of Trust in Social Organizations

Trust in Organization

Competence Social Value Similarity

4.2 Measurement

A key issue is how one can measure trust. Psychometric research into trusthas suggested that direct questions of trust engage two components: socialvalue similarity and competence (Cvetkovich & Nakayachi 2007, Nakayachi& Cvetkovich 2010). We do not have responses directly related to socialvalue similarity perceptions of DHS. Instead of choosing the general trustquestions (which combines these two effects without a method for separatingout the components), we have elected to use the competence component oftrust given the choice of survey question available to us. These tests will thenfocus on the competence component and the results may not be generalizableto the social value similarity component – which we leave to future research.The specific question wording is presented in appendix one.

The competence question results in responses along an 11 point scale.The raw response distribution is illustrated in the first panel of Figure 2.

This number of categories and the skew present in the distribution makethe data difficult to analyze. We have rescaled the variable for analysis bycollapsing the original scale from zero to four; as well as collapsing nine andten. The recoded version of the competency assessments are illustrated inthe second panel of Figure 2.

The remainder of the measures are relatively straight forward2. The de-scriptive statistics for all variables are offered in 1.3

2For more information on the approach to measuring the independent variables, Ap-pendix one includes coding and survey measurement information from the survey

3There were significant missing values, particularly in the income variable. The typicalsolution, listwise deletion, can lead to biased results (Rubin 1987, King, Honaker, Joseph& Scheve 2001). To avoid these biases, we have employed multiple imputation to generatepredicted values for missing data points of independent variables resulting in five simulateddatasets. Table 2 reports the results based on these imputed data sets. We replicated theanalysis without imputation to ensure that there were not remarkable differences. The

10

Table 1: Initial Regression Results

Variable x st. dev. min. max. # missingCompetence 2.69 1.02 0 4 30

HS Policy 2.90 1.06 0 4 4Salience

Party(R) .291 .454 0 1 52

Party(I) .405 .491 0 1 52

Education 15 2.14 11 18 6

Religiosity .452 .480 0 1 14

Conservative .378 .485 0 1 44

Moderate .459 .497 0 1 44

Gen. Gov’t Trust 1.16 .870 0 3 13

Age 55.2 15.3 19 93 27

Income 61 31.5 5 100 228

# of Kids .542 1.03 0 9 8

Gender .520 .500 0 1 0

Black .876 .329 0 1 27

(N = 924)Missing values imputed (except for Competence) for each analysis

11

Figure 2: The Distribution of Competence of DHS

4.3 Generalized Ordered Logit Models

The five-point scale as a dependent variable presents some challenges fortraditional regression analysis. Simple OLS regression would impose an as-sumption on the dependent variable that the categories were equally spacedand that fractional responses were possible. This was clearly not the case.We have instead opted to use ordered logit regression to account for theordered but not continuous nature of the dependent variable (McKelvey &Zavoina 1975).

This model does require a parallel regression assumption (Long 1997).This assumption holds that the impact of any variable is constant across theentire range of the dependent variable.4 For example, the ordered logit modelassumes that the impact of being a moderate is the same in differentiating0 from higher levels of competence as it does from the maximum and lower

differences between the imputed and non-imputed results were driven in all cases by thereduced sample size of the non-imputed data set – particularly when including the Incomevariable. This was clear because the differences were driven by the standard errors inproportion to missing observations.

4As would simple OLS regression, for that matter.

12

values of trust. Whether this assumption holds is testable with a Brant Test.We will report these tests as well as relax the assumption using generalizedordered logit in situations where variables fail the Brant test (Williams 2006).The generalized ordered logit model operates in a manner similar to a local(loess) regression model wherein the impact of an explanatory variable mayvary of the range of the dependent variable. In the case of generalized orderedlogit, the dependent variable is only observed in discrete levels rather thancontinuously – thus making a local (loess) estimator inappropriate.

5 Results

The results from the various regression models are quite complex. The nextsection discusses the direct interpretation of the coefficients, hypothesis tests,and model fit. The subsequent section will present the substantive interpre-tation of effect sizes for key variables.

5.1 Regression Results

Table 2 presents the coefficient estimates and model fit diagnostics for ourthree models.5 The first model (OLOGIT 1) presents the initial test ofhypotheses one through four. This model most closely resembles the specifi-cation strategy common in previous research into political trust. Hypothesisone does not find strong support in this initial model. The effect of reli-giosity is positive but does not meet the traditional standard of a z-score of1.96 (corresponding to a p-value of .05). The only political attitude variablesignificant (partially supporting hypothesis three) is self-identified partisanindependence. Independents have significantly lower assessments of the com-petence of DHS than democratic party identifiers. Interestingly, this variableis significant while republican party identifiers are not significantly differentthan the baseline category (democratic party identifiers). The strongest sup-port comes for hypothesis four. Reported attention to homeland security is astrong predictor of competence assessments significantly increasing expectedassessment levels. These results provide a simple test, though not unlikemany models of political trust, establishing the importance of including pol-

5Table 2 only presents the results related to our core variables. The results for thecontrol variables (listed at the bottom of the table) are presented in appendix two.

13

icy domain specific variables – particularly, attention to the policy areas – inmodels of assessments of a specific agency.

The overall model fit is difficult to assess with a categorical dependentvariable. McFadden’s R2 is commonly reported but there are no clear guide-lines as to what counts as an acceptable model fit. Here the value of approxi-mately .03 represents a ratio of the unconstrained and the modeled likelihoodratios. This value tends to fall as the sample size increases making compar-isons to other models with different numbers of observations difficult. Thecount R2 has a more natural interpretation. For each observation, the levelof trust with the highest predicted probability from the model is comparedto the observed value. The count R2 represent the percentage of correct pre-dictions. The basic model correctly predicts the observed value 29.5% of thetime.

It is possible that these results are driven by general trust assessments –with assessments of DHS being epiphenomenal. This would not explain thewidely varied assessments of agencies reported elsewhere, but it is possiblethat differences in agencies are not systematically related to the factors out-lined in hypotheses one through three. To test this possibility, we estimateda model that included generalized trust as an explanatory variable. As pre-dicted in hypothesis five, generalized trust is a significant predictor of specificassessments of DHS. This is not surprising. What may be surprising is thepersistence of other effects in the presence of the generalized trust controlvariable. Religiosity becomes significant providing evidence for hypothesisone. Respondents who reported that they attend a religious service at leastonce a week have higher expected levels of reported competence of DHS.Reported attention to homeland security is still strongly significant as well,not surprising given the domain specific nature of this variable. The measureof generalized trust is an index of trust assessments of political institutions(Congress and the president) with strong partisan identification. Predictably,the partisan identification variables no longer have a significant direct effect.However, the filtering of the partisanship effect through generalized politicaltrust has increased the significance of the ideology measures (of self-identifiedconservative and moderate ideology) – though not quite to traditional levels.The OLOGIT 2 model offers evidence in support of hypotheses one, three(partially), four, and five.

The fit of the second ordered logit model is better with an improved Mc-Fadden’s R2 and an improved count R2. The magnitude of the improvementfor McFadden’s R2 is difficult to interpret directly – though there is an im-

14

provement. The count R2 has the directly interpretable meaning of correctlypredicting 31.3% of the cases.

Table 2: Initial Regression Results

Variable OLOGIT 1 OLOGIT 2 GOLOGIT

Religiosity .238 (1.92) .251 (2.01) .623 (2.90)

Party(R) -.197 (-1.03) .1747 (.858) .155 (.760)

Party(I) -.504 (-3.21) -.237 (-1.44) -.238 (-1.45)

Conservative .040 (.182) .448 (1.93) 1.28 (3.70)

Moderate .235 (1.23) .356 (1.86) .836 (3.02)

HS Policy .318 (5.23) .363 (5.85) .355 (5.73)Salience

Gen. Gov’t Trust .544 (5.66) .854 (5.66)

McFadden’s R2 .0315 .0431 .0528Count R2 .295 .313 unavailable

Missing values are imputed (m=5) for n=894Z-statistics in parentheses (with Rubin’s correction for imputation)

Controls included for: gender, age, education, # of children, and income.

Brant tests following the OLOGIT 2 model revealed significant viola-tions of the parallel regression assumption. For several variables (religiosity,moderate ideology, conservative ideology, and generalized political trust), theeffect of the variable was significantly different at different levels of the de-pendent variable. This compels us to re-estimate the model with generalizedordered logit to allow these variables to have different coefficients across lev-els of the dependent variable. Each of the variables that failed the Brant

15

test were allowed to have varying coefficients. The generalized ordered logitmodel results are reported, in part, in the third column of Table 2. It isimportant to note that for the variables we allowed to have varying coeffi-cients (religiosity, moderate ideology, conservative ideology, and generaliedpolitical trust) the coefficient reported on Table 2 only represents the effectat the lowest level of the dependent variable (separating the lowest level fromhigher levels).

The baseline coefficient now only represent the effect at the lowest levelsof the dependent variable (between competence ratings of 0 and 1). At theselowest levels of reported competence assessment, religiosity, conservative ide-ology, and moderate ideology all have a significant positive effect on expectedcompetence assessments. The Brant test suggests that the effects of the vari-ables may change at higher levels of competence assessment. Table 3 providesthe gamma values describing how these estimated coefficients change as oneconsiders higher values of the dependent variable. In each case (though onlymarginaly significant in the case of religiosity), the coefficient of the variablesis smaller at higher levels of the dependent variable. In all of these cases,we see a prophylactic effect. Increases in each factor reduce the probabilityof observing the lowest level of competency assessment. However, this effectis largely muted at higher levels of reported trust. As an example, the γ2for religiosity means that the net effect of religiosity on distinguishing theprobability of observing the two lowest levels of competence from the higherlevels is .623− .741. The result is not statistically distinguishable from zero.What distinguishes this prophylactic effect from a traditional effect is thatfor each variable, the improvement in reported competence is only operatingat preventing the lowest levels. There is not nearly as strong a result inpromoting the highest levels (from the next lower level). The effect of home-land security policy salience passess the Brant test and has an effect (andsignificance) similar to what we found in the previous models. Here againhypothesis three finds strong evidence along with the persistence of the effectof policy salience included as part of hypotheses five. In this fully specifiedmodel, four of the five hypotheses (again excepting the partisan identificationmeasures that are so closely related to the measure of generalized politicaltrust) receive support.

The fit of the generalized ordered logit model is better with an improvedMcFadden’s R2. The improvement is not as dramatic as between the twoordered logit models but noticeable. In this case, we can not calculate thecount R2 because the generalized ordered logit algorithm is inconsistent with

16

Table 3: GOLOGIT Gamma Values

Variable Religiosity Moderate Conservative Trust

γ(2) -.741(-2.35) -.500 (-2.18) -.887(-3.14) -.492(-3.10)

γ(3) -.392(-1.80) -.492(-1.84) -.889(-2.73) -.361(-2.31)

γ(4) -.407(-1.64) -.741(-2.35) -1.18(-3.06) -.279(-1.619)

the post-estimation algorithm to assess count R2.

5.2 Simulation Results

While the direction and statistical significance allowed us to test our hy-potheses, the data provide more information about the substantive impactof the key variables. While in a linear regression model the coefficient has adirect and simple interpretation; this is not the case in non-linear models likeordered logit and generalized ordered logit. Instead, we provide illustrationsof the expected distribution of competency assessments for key illustrativecases based on the results of the GOLOGIT model.6 Figure 3 compares thetypical case to one where the respondent reports attending religious servicesweekly (or more often). The expected value (mean) changes from 1.8 to 2.0.The figure illustrates that the changes are not as simple as a constant changeacross the values of the dependent variable. Instead, the changes are concen-

6We used prvalue to generated predicted probabilities for each possible value of com-petence given the median value of each variable – with the exception that the ideologybaseline is “liberal” to allow for greater contrast of “moderates” and ”conservatives”. Thisis the “typical” case. For dichotomous variables, we compare the “typical” case to thepredicted values for a case with all variables at their median except for the variable namedbelow the figure. For variables with more than two values, we illustrate the predictedprobabilities for each value of the key variable with all other variables at their medianvalues. We then these predicted probability to determine the expected distribution of1000 responses for each set of characteristics – choosing 1000 to avoid rounding errors.The expected distributions provide a clear visualization of the changes in the expectedresponses as we change each variable in turn.

17

trated in the lower parts of the distribution with a remarkable change in theprobability of the lowest level of competence (0). This figure illustrates theprophylactic effect discussed in interpreting the coefficients of the regressionmodel.

Figure 3: Simulated Distribution of Competency for Typical Case and Reli-giosity

Figures 5 and 4 similarly illustrate the impact of the moderate and con-servative ideology, respectively. Moderates have an expected value of 2.1compared to the typical case of 1.8. Similarly, conservative respondents havean expected value of 2.2.

It is interesting that self-identified moderates and conservatives havelarger competence assessments than the omitted category (liberals) in anera where a democratic president is in the White House. Future researchshould investigate whether this relationship holds up with regards to agen-cies more closely associated with liberal policies (e.g. Health and HumanServices). In each of these cases we again see that the largest changes are inthe lowest categories of competence assessments. Again the figures illustratea prophylactic effect.

18

Figure 4: Simulated Distribution of Competency for Typical Case and Mod-erate Ideology

We focus finally on the impact of different levels of attention to issues ofhomeland security on competence assessment. Substantively, the expectedvalue of competence assessment start at 1.0 for respondents reporting a lowlevel of attention to homeland security and increase to 1.2, 1.5, 1.8, and 2.1as we increase the level of attention. The generalized ordered logit revealedthat the impact of attention is uniform across levels of the dependent vari-able. Figure 6 illustrates the change in distribution with reductions in theprobability of low values and increases in the probability of high values asattention increases. The case of attention illustrates how a variable can af-fect competency assessment without the prophylactic pattern observed in thecases of religiosity and ideology.

6 Conclusion

The previous section illustrated that individual assessments of DHS dependcritically on individual characteristics. Various characteristics such as demo-graphics, political attitudes, and the personal salience of the policy area have

19

Figure 5: Simulated Distribution of Competency for Typical Case and Con-servative Ideology

substantively significant impacts on the expected levels as assessed compe-tence – our chosen measure of trust. This has a number of implications forthe study of trust in administrative agencies.

First, trust in administrative agencies is not driven entirely by the behav-ior of the agency or macro-political changes in political attitudes. There is agreat deal of variation in the assessments of a given agency across the public.An important implication of this variation is that any attempt to improvethe trust in an agency should be grounded in an awareness of the segmentsof society where there is a trust gap. In the example of the DHS, it is clearthat higher levels of salience for homeland security significantly increases theassessed competence of DHS (which we take to be a specific measure of trustin DHS). An efficient strategy to increase trust in DHS might involve aneffort to increase the salience of homeland security.

Second, the individual factors related to trust in DHS were different thanthose often included in related research. Within the fiend of public adminis-tration, measures of political attitudes are seldom included. The significantinfluence of these attitudes in our models suggest that this is an important

20

Figure 6: Simulated Distribution of Competency for Different Levels of At-tention to Homeland Security

omission. Within political science models, religiosity is seldom included butis significant here. This omission may again be important.

Third, we have focused on the evaluation of a specific agency rather thana general question of trust in government. The robustness of the findingswhen including a control for the respondent’s general trust in governmentillustrates that there are factors specific to DHS that are not fully explainedby generalized political trust. The results suggest that models of trust forDHS need to be specific to the context of homeland security. This suggeststhat a great deal more work lies ahead for building models of trust for specificadministrative agencies. While research into generalized political trust is stillimportant, the results here suggest that such research leaves a great deal ofinteresting variation ignored – variation by agencies.

Fourth, the complexity required in the statistical model revealed the lim-itations of the most common approach to models with ordered dependentvariables. If we had stopped with the ordered logit model, we would havecreated sensible results. However, these results would have been partial –and partially misleading. Moving to generalized ordered logit revealed the

21

prophylactic effect of religiosity and political attitudes. These results sug-gest that greater attention should be paid to the seemingly innocuous parallelregression assumption.

Future research can go in two directions. First, it is clear that a deeperunderstanding of trust in specific agencies will require additional work onthe role of policy domain factors, other individual characteristics, and thebehavior of the agencies themselves. The work here reveals the merit of suchfocused research but does not exhaust that questions that we must addressrelated to the evaluation of specific agencies. Second, we may start to lookfor patterns between agencies. While current research typically lumps allagencies together and suggests that trust is a shared characteristic. Greaterunderstanding of the specific factors related to specific agencies may revealclusters of similar agencies. Do individuals evaluate defense agencies similarly– even if they evaluate social welfare agencies differently? Is DHS a defenseagency? Greater understanding of the specific agency models can facilitatethe emergence of groupings of agencies.

22

Appendix One – Survey Details

The phone survey included a wide range of questions related to nuclear se-curity issues, assessments of specific administration agencies, and some basicdemographic information. The questions that serve as the basis for the dataused in the models reported in this article are presented here.

Variable – Competence

I am going to read a list of groups that make decisions that affecthomeland security. Using a scale of 0 to 10, where 0 means not atall competent, and 10 means completely competent, how wouldyou rate the competence of each group to make decisions abouthomeland security policy?

• US Department of Homeland Security

• US Department of Defense

• US Customs and Border Protection Agency

• US Department of Energy

• US Department of State

• Central Intelligence of Agency

• National Nuclear Security Administration

• US Coast Guard

• Federal Bureau of Investigations

• US Department of Agriculture

• Don’t Know

The competence measured used the response for the US Department ofHomeland Security. The order of the list was randomized when adminis-tered over the phone.

Variable – Homeland Security Salience

23

On a scale from 0 to 10, with 0 meaning no attention and 10meaning much attention, how much attention do you personallypay to issues of national security and terrorism?

This variable was scaled: 0–1=1, 2–3=1, 4–6=2, 7–8=3, 9–10=4

Variable – Party Identification

Generally speaking, do you think of yourself as a Democrat, Re-publican, or Independent?

• Democrat

• Republican

• Independent

• Don’t know

Party identification variables dichotomously as 1 as two variables (Republi-can, Independent)

Variable – Ideology

Which of the following categories best describes your politicalviews?

• Strongly Liberal

• Liberal

• Slightly Liberal

• Middle of the Road

• Slightly Conservative

• Conservative, or

• Strongly Conservative

24

• Other(Specify)

Party identification variables dichotomously as 1 as two variables: Conserva-tive and Strongly Conservative = Conservative; Slightly Conservative, Mid-dle of the Road, and Slightly Liberal = Moderate.

Variable – Generalized Political TrustGeneralized Political Trust is an index of responses to two questions.

Thinking about the way Barack Obama is doing his job as Pres-ident, do you strongly approve, approve, disapprove, or stronglydisapprove?

Thinking about the way Congress is doing its job, do you stronglyapprove, approve, disapprove, or strongly disapprove?

Both questions used the following response options:

1. Strongly Approve

2. Approve

3. Disapprove

4. Strongly Disapprove

5. Don’t Know

The responses were added and direction was reversed so that the index rep-resents increasing trust.

Variable – Education

What is the highest level of education you have completed?

• Elementary or Some High School

• High School Graduate / GED

• Trade or Vocational Certification

25

• Some College / Associates Degree

• College Graduate

• Post-graduate Degree

Variable – Age

How old are you? [in years]

Variable – Race

From the following options, do you consider yourself to be:

• Black, or African-American

• White

• Asian

• American Indian of Alaska Native

• Native Hawaiian or other Pacific Islander

• Other

We coded ”white” as a dichotomous variable which equals ”1” when respon-dents indicated ”White” – and ”0” otherwise.

Variable – # of Children

How many children under the age of 18 live in your household?

Variable – Income

What was the estimated annual income for your household for2008?

• Less than $10,000

26

• 10 to $20,000

• 21 to $30,000

• 31 to $40,000

• 41 to $50,000

• 51 to $60,000

• 61 to $70,000

• 71 to $80,000

• 81 to $90,000

• 91 to $100,000

• More than $100,000

For imputation purposes income was recoded taking the midpoint of eachrange (e.g. 10 to $20,000 was recoded 15,000). After imputation, the valueswere truncated into the same the same ranges set at the midpoint of theoriginal range (e.g. any value between 10,000 and 20,000 was recoded 15).

Variable – Religiosity

Did you attend church, synagogue, mosque, or any other type ofreligious service in the last 7 days?

• Yes

• No

Variable – Gender

As part of the survey, I am required to ask: are you male orfemale?

• Male

• Female

We coded gender as a dichotomous variable which equals ”1” when respon-dents indicated ”Female” – and ”0” otherwise.

27

Appendix Two – Control Variables for Regres-

sion Models

Coming soon.

28

References

Alford, J.R. 2001. We’re All in This Together The Decline of Trust in Gov-ernment, 1958-1996. Cambridge University Press.

Bafumi, J. & R.Y. Shapiro. 2009. “A New Partisan Voter.” The Journal ofPolitics 71(01):1–24.

Baumgartner, J.C., P.L. Francia & J.S. Morris. 2008. “A Clash of Civiliza-tions? The Influence of Religion on Public Opinion of US Foreign Policyin the Middle East.” Political Research Quarterly 61(2):171.

Berkman, M.B. & E. Plutzer. 2009. “Scientific Expertise and the CultureWar: Public Opinion and the Teaching of Evolution in the AmericanStates.” Perspectives on Politics 7(03):485–499.

Black, G.S. & B.D. Black. 1994. The Politics of American Discontent: Howa New Party Can Make Democracy Work Again. Wiley.

Blendon, R.J., J.M. Benson, R. Morin, D.E. Altman, M. Brodie, M. Brossard& M. James. 1997. “Changing Attitudes in America.” Why People DontTrust Government pp. 205–16.

Bok, D. 1997. “Measuring the Performance of Government.” Why PeopleDont Trust Government Eds. S. Joseph Nye, P.D. Zelikow, and D.C.King pp. 55–76.

Bolzendahl, C. & C. Brooks. 2005. “Polarization, Secularization, or Differ-ences as Usual? The Denominational Cleavage in US Social Attitudessince the 1970s.” Sociological Quarterly 46(1):47–78.

Brewer, P.R. & L. Sigelman. 2002. “Trust in Government: Personal TiesThat Bind?” Social Science Quarterly 83(2):624–631.

Brooks, C. 2002. “Religious Influence and the Politics of Family DeclineConcern: Trends, Sources, and US Political Behavior.” American Soci-ological Review pp. 191–211.

Caldeira, G.A. & J.L. Gibson. 1992. “The Etiology of Public Support for theSupreme Court.” American Journal of Political Science pp. 635–664.

29

Campbell, D.E. & J.Q. Monson. 2008. “The Religion Card: Gay Marriageand the 2004 Presidential Election.” Public Opinion Quarterly .

Catterberg, G. & A. Moreno. 2006. “The Individual Bases of Political Trust:Trends in New and Established Democracies.” International Journal ofPublic Opinion Research 18(1):31.

Chanley, V.A. 2002. “Trust in Government in the Aftermath of 9/11: De-terminants and Consequences.” Political Psychology 23(3):469–483.

Chanley, V.A., T.J. Rudolph & W.M. Rahn. 2000. “The Origins and Conse-quences of Public Trust in Government: A Time Series Analysis.” PublicOpinion Quarterly 64(3):239.

Cheema, G.S. 2005. Building Democratic Institutions: Governance Reformin Developing Countries. Kumarian Press.

Citrin, J. & D.P. Green. 1986. “Presidential Leadership and the Resurgenceof Trust in Government.” British Journal of Political Science 16(4):431–53.

Citrin, J. & S. Luks. 2001. “Political Trust Revisited: deja vu all over again?”What Is It about Government that Americans Dislike pp. 9–27.

Cook, T.E. & P. Gronke. 2005. “The Skeptical American: Revisiting theMeanings of Trust in Government and Confidence in Institutions.” Jour-nal of Politics 67(3):784–803.

Craig, S.C. 1993. The Malevolent Leaders: Popular Discontent in America.Westview Press Boulder, Colorado.

Cvetkovich, G. & K. Nakayachi. 2007. “Trust in a High-concern Risk Con-troversy: A Comparison of Three Concepts.” Journal of Risk Research10(2):223–237.

Dahl, R.A. 1971. Polyarchy: Participation and Opposition. Vol. 54 YaleUniversity Press.

Easton, D. 1965. A Systems Analysis of Political Life. Wiley.

Easton, D. 1975. “A Re-Assessment of the Concept of Political Support.”British Journal of Political Science 5(4):435–457.

30

Erber, R. & R.R. Lau. 1990. “Political Cynicism Revisited: An Information-processing Reconciliation of Policy-based and Incumbency-based Inter-pretations of Changes in Trust in Government.” American Journal ofPolitical Science pp. 236–253.

Feldman, Stanley. 1983. “The Measure and Meaning of Trust in Govern-ment.” Political Methology 9:341–54.

Fiorina, M.P. & S.J. Abrams. 2008. “Political Polarization in the AmericanPublic.” Annu. Rev. Polit. Sci. 11:563–588.

Froese, P. & F.C. Mencken. 2009. “A US Holy War? The Effects of Religionon Iraq War Policy Attitudes*.” Social Science Quarterly 90(1):103–116.

Garment, S. 1991. Scandal: The Crisis of Mistrust in American Politics.Random House.

Gershtenson, J., J. Ladewig & D.L. Plane. 2006. “Parties, InstitutionalControl, and Trust in Government.” Social science quarterly 87(4):882–902.

Gibson, J.L., G.A. Caldeira & L.K. Spence. 2003. “Measuring Attitudes to-ward the United States Supreme Court.” American Journal of PoliticalScience 47(2):354–367.

Hetherington, M.J. 1998. “The Political Relevance of Political Trust.” Amer-ican Political Science Review pp. 791–808.

Hetherington, M.J. 1999. “The Effect of Political Trust on the PresidentialVote, 1968-96.” American Political Science Review pp. 311–326.

Hetherington, M.J. 2005. Why Trust Matters: Declining Political Trust andthe Demise of American Liberalism. Princeton University Press.

Holbrook, A., J.A. Krosnick & A. Pfent. 2007. The Causes And ConsequencesOf Response Rates In Surveys By The News Media And GovernmentContractor Survey Research Firms. In Advances in Telephone SurveyMethodology, ed. et al. James M. Lepkowski. Wiley.

Houston, D.J., P.K. Freeman & D.L. Feldman. 2008. “How Naked is thePublic Square? Religion, Public Service, and Implications for PublicAdministration.” Public Administration Review 68(3):428–444.

31

Jennings, M.K. 1998. “Political Trust and the Roots of Devolution.” Trustand Governance 1:218–44.

Jones, J. M. 2006. “Trust in Government: Declining, Near Lows for thePast Decade; Less than Half Express Trust and confidence in ExecutiveBranch.” The Gallup Poll News Service. Princeton, New Jersey.

Keele, L. 2005. “The Authorities Really Do Matter: Party Control and Trustin Government.” Journal of Politics 67(3):873–886.

Keele, L. 2007. “Social Capital and the Dynamics of Trust in Government.”American Journal of Political Science 51(2):241–254.

Keeter, S., C. Kennedy, M. Dimock, J. Best & P. Craighill. 2006. “Gaugingthe impact of Growing Nonresponse on Estimates from a National RDDTelephone Survey.” Public Opinion Quarterly 70(5):759.

King, G., J. Honaker, A. Joseph & K. Scheve. 2001. “Analyzing IncompletePolitical Science Data: An Alternative Algorithm for Multiple Imputa-tion.” American Political Science Review 95(1):49–70.

Lawrence, R.Z. 1997. “Is It Really the Economy, Stupid?” Why People DontTrust Government pp. 111–32.

Levi, M. 1997. Consent, Dissent, and Patriotism. Cambridge UniversityPress.

Lipset, S.M. & W. Schneider. 1983. The Confidence Gap: Business, Labor,and Government in the Public Mind. Free Press New York.

Long, J.S. 1997. Regression Models for Categorical and Limited DependentVariables. Vol. 7 Sage Publications, Inc.

Mansbridge, J. 1997. “Social and Cultural Causes of Dissatisfaction with USGovernment.” Why People Dont Trust Government pp. 133–153.

McKelvey, R.D. & W. Zavoina. 1975. “A Statistical Model for the Analysis ofOrdinal Level Dependent Variables.” Journal of Mathematical Sociology4(1):103–120.

Merkle, D. & M. Edelman. 2002. Nonresponse in Exit Polls: A Comprehen-sive Analysis. In Survey Nonresponse. Wiley pp. 243–58.

32

Miller, A.H. & S.A. Borrelli. 1991. “Confidence in Government During the1980s.” American Politics Research 19(2):147.

Miller, W.E. 1991. “Party Identification, Realignment, and Party Voting:Back to the Basics.” The American Political Science Review pp. 557–568.

Morgeson, F.V., D. VanAmburg & S. Mithas. 2011. “Misplaced Trust? Ex-ploring the Structure of the E-Government-Citizen Trust Relationship.”Journal of Public Administration Research and Theory 21(2):257.

Nakayachi, K. & G. Cvetkovich. 2010. “Public Trust in Government Con-cerning Tobacco Control in Japan.” Risk Analysis 30(1):143–152.

Orren, G. 1997. “Fall from Grace: The Publics Loss of Faith in Government.”Why People Dont Trust Government pp. 77–107.

Pew Research Center. 1998. “Deconstructing Distrust: How Americans ViewGovernment.”.

Pew Research Center. 2010. “Distrust, Discontent, Anger and Partisan Ran-cor: The People and Their Government.” Available at http://people-press.org/report/606/trust-in-government.

Putnam, R.D. 1993. Making Democracy Work. Princeton University Press.

Putnam, R.D. 2001. Bowling Alone: The Collapse and Revival of AmericanCommunity. Simon and Schuster.

Rubin, D.B. 1987. Multiple Imputation for Nonresponse in Surveys. Vol. 17John Wiley & Sons Inc.

Rudolph, T.J. 2009. “Political Trust, Ideology, and Public Support for TaxCuts.” Public Opinion Quarterly 73(1):144.

Rudolph, T.J. & J. Evans. 2005. “Political Trust, Ideology, and Public Sup-port for Government Spending.” American Journal of Political Sciencepp. 660–671.

Scholz, J.T. & M. Lubell. 1998. “Trust and Taxpaying: Testing the HeuristicApproach to Collective Action.” American Journal of Political Sciencepp. 398–417.

33

Scholz, J.T. & N. Pinney. 1995. “Duty, Fear, and Tax Compliance: TheHeuristic Basis of Citizenship Behavior.” American Journal of PoliticalScience pp. 490–512.

Sherkat, D.E. 2011. “Society Without God: What the Least Religious Na-tions Can Tell Us about Contentment.” Contemporary Sociology: AJournal of Reviews 40(4):499–500.

Statacorp. 2009. “Stata Statistical Software: Release 11.” College Station,TX: Statecorp LP.

Stokes, C.E. & C.G. Ellison. 2010. “Religion and Attitudes Toward DivorceLaws Among US Adults.” Journal of Family Issues 31(10):1279.

Stoutenborough, J.W. & D.P. Haider-Markel. 2008. “Public Confidence inthe US Supreme Court: A New Look at the Impact of Court Decisions.”The Social Science Journal 45(1):28–47.

Tolbert, C.J. & K. Mossberger. 2006. “The Effects of E-Government onTrust and Confidence in Government.” Public Administration Review66(3):354–369.

Tyler, T.R. 1990. “Why People Obey the Law: Procedural Justice, Legiti-macy, and Compliance.”.

Tyler, T.R. & P. Degoey. 1995. “Collective Restraint in Social Dilemmas:Procedural Justice and Social Identification Effects on Support for Au-thorities.” Journal of Personality and Social Psychology 69(3):482.

Welch, E.W., C.C. Hinnant & M.J. Moon. 2005. “Linking Citizen Satisfac-tion with E-government and Trust in Government.” Journal of PublicAdministration Research and Theory 15(3):371.

Williams, J.T. 1985. “Systemic Influences on Political Trust: The Impor-tance of Perceived Institutional Performance.” Political Methodology11(1-2):125–42.

Williams, R. 2006. “Generalized Ordered Logit/Partial Proportional OddsModels for Ordinal Dependent Variables.” Stata Journal 6(1):58–82.

34