Strong Impact of Interviewers on Respondents’ Political Choice

Abstract

Despite much literature on interviewer effects, limited attention has been paid to party preference surveys, although the effect is expectedly strong in this field. This article analyzes interviewer effects in a face-to-face political survey. Specifically, we are interested in whether the interviewer’s own party preference has an effect on the respondent’s party choice. We used cross-classified two-level logistic regression models with median odds ratio as effect size. We found four main results: (1) Place of residence significantly affects political preference, but interviewers do so to the same degree; (2) the size of these effects is comparable to those of demographic characteristics of the respondent; (3) interviewers’ political preference has an effect and it does not disappear once controlled for obvious interviewer characteristics; and (4) the impact of political preference is such that respondents tend to have a preference similar to that of their interviewers.

The problem of interviewer effects has long been known in survey research literature (see Fowler and Mangione [1990] and Hyman [1954] as two classics). Numerous attempts have been made to measure interviewer effects, and the past two decades have seen a unification of the statistical methodology of measurement with the advent of multilevel statistical models (P. Davis and Scott 1995; Hox 1994; O’Muircheartaigh and Campanelli 1998; Pickery and Loosveldt 2002; Schnell and Kreuter 2005).

To the best of our knowledge, no study has examined the effect in party preference surveys—even though party preference is among those sensitive, nonfactual, hard-to-answer topics where interviewer effects are probably rather strong. A lot of research has been carried out, however, on strong interviewer effects on other political topics (e.g., D. W. Davis and Silver [2003] on political knowledge; Lipps and Lutz [2010] on the politics of Switzerland; and Pickery and Loosveldt [2002] on nonresponse to party preference).

Various explanations have been proposed concerning the source of interviewer effects including the nonstandardized ways of formulating and asking questions, recoding answers, probing inadequate answers, or handling interpersonal aspects (Fowler and Mangione 1990). Interviewer effects are generated by sensitive, nonfactual, difficult, or open items. According to the explanations, the questions here are more open to interpretation, so the respondent is more likely to ask the interviewer for help (e.g., Fowler and Mangione 1990; Schnell and Kreuter 2005). Some authors (e.g., West and Blom 2016) point out that interviewer effects can affect not only the content of the responses (measurement error) but also willingness to respond (nonresponse error).

Apart from their behavior, visible traits and other personal characteristics of the interviewers may also influence the responses, especially if the topic of the survey can be linked with these characteristics (e.g., Hyman 1954). The general explanation for this is known in survey literature as social desirability bias (e.g., Berinsky 1999). This says that respondents answer some sensitive questions in such a way as to comply with what they think society find more desirable, even if they hold only vague or no preferences on the issue or have a different opinion. The phenomenon has been linked with the measurement of political attitudes (e.g., Lipps and Lutz 2010), thus it can work specifically when surveying political party preferences. Conformity with the majority can, on the one hand, result in the overestimation of fixed party preference (Nishizawa and Kuriyama 2016) because of the norm that responsible citizens in a democracy are expected to participate. On the other hand, political surveys are special in that respondents state their preference knowing about the preference of the majority (which is regularly published). It can be argued that the difference between private and public political opinion is due to the fear of isolation—the fear of being different from others (e.g., Shoemaker et al. 2000).

Social desirability bias can explain the impact of the interviewer’s opinion as well; since in the interview situation, it is the interviewer with whom the respondent would like to build a positive self-image (e.g., Krysan and Couper 2003). Even if interviewers do not disclose their attitudes, there must be nonverbal channels that signpost them for the respondent (Lipps and Lutz 2010). Pickery and Loosveldt (2002) used the political interest of the interviewer as a predictor and found it significant in some cases. Katz’s (1942) classic experiment and Lipps and Lutz (2010) found that respondents are more likely to have a political opinion similar to the opinion of the interviewer.

In the light of all this, in the present study, we focused on interviewers’ own political preferences and those characteristics that could be manifest for respondents from interviewers’ appearance and behavior.

Another reason to measure interviewer effects is the challenge presented by its interrelatedness with geographical context. As we know from survey sampling literature (e.g., Kish 1995), there is usually a positive correlation among characteristics for people belonging to the same area clusters. This positive correlation is also seen for political questions. As Burbank (1997) points out, the explanations for this center around the environmental filtering of information concerning political matters. The question of the impact of place of residence on party preference is addressed by the field of electoral geography, which has been attracting increasing attention in recent years. However, the studies that examine the impact of geographic context on survey data can be misleading if they ignore interviewer effects, since the assignment of respondents to interviewers is, in most cases, usually not random. That is, they don’t use interpenetrating sampling that would make it possible to separate settlement and interviewer effects (see, e.g., Biemer and Stokes 1985). Instead, those living in the same neighborhood are likely to be assigned to the same interviewer. Thus, by ignoring interviewers’ effect, one ends up overestimating the effect of geographical context.

Only a few studies have estimated the relative impact of interviewer and geographical context (see P. Davis and Scott 1995; O’Muircheartaigh and Campanelli 1998; Schnell and Kreuter 2005; Turner et al. 2015). Each of these surveys were based on face-to-face surveys, with O’Muircheartaigh and Campanelli (1998) and Schnell and Kreuter (2005) using interpenetrated design. They all used a multilevel regression model for separating the two effects. They detected significant interviewer effect, whose extent was commensurate with or exceeded the effect of the place of residence for nonfactual questions. We haven’t found any studies measuring the two types of effects at the same time for political attitudes.

In our study, we tried to measure the interviewers’ effect on self-reported party preference, with special attention to the effect of the interviewer’s own political preference. We aimed to calculate an effect size that makes the interviewer effects directly comparable with the effect of other known determinants. Also, we tried to separate interviewer effects from the effect of place of residence, using multilevel regression models. Our research was made possible by an exceptionally informative database that of a Hungarian large-sample survey,¹ which included questionnaires filled in by the interviewers themselves.

Data

The data we used come from a series of (non-follow-up) surveys carried out on a monthly basis in 2010. For our analysis, we used the answers to the question: “Which party would you vote for if the elections were held this Sunday?” Because of sample size limitations, we restricted our analysis to three parties: Fidesz–Hungarian Civic Union (Fidesz), magyar szocialista párt (MSZP) (Hungarian Socialist Party), and the Movement for a Better Hungary (Jobbik).

Interpretation of the results necessitates some understanding of the Hungarian political context. In April 2010, there were parliamentary elections in Hungary; the Fidesz, a right-wing conservative party allied with the kereszténydemokrata néppárt (KDNP) (Christian Democratic People's Party), won the elections. Fidesz’s landslide victory was a result of massive dissatisfaction with MSZP (see, e.g., Lansford 2014), which had been in power since 2002. Two new parties emerged and entered the parliament, including Jobbik, an extreme right-wing party. The composition of the parliament became as follows: Fidesz-KDNP 68.1%, MSZP 15.3%, Jobbik 12.2%, and lehet más a politika (LMP) (politics can be different) 4.15%.

The interviews were conducted face to face. In May, during their training, interviewers also filled out the questionnaire. Most settlements were not visited every month, and there was some fluctuation in the group of the interviewers over the months as well. Since political preferences can change over time, this fluctuation could lead us to incorrectly overestimate interviewer effects. To avoid this, the month when the interview was conducted was included in every model. In our analysis, we only included cases with valid values for all variables—which yielded about 5,000 respondents, from a total of 147 settlements and with 73 interviewers.

Method

The respondents were cross-classified by interviewers and settlements (i.e., there were settlements with several interviewers and there were interviewers assigned to several settlements). Thus, we used cross-classified multilevel regression models (in the context of interviewer effects, see P. Davis and Scott 1995; Lipps and Lutz 2010; O’Muircheartaigh and Campanelli 1998; Schnell and Kreuter 2005; Turner et al. 2015). Our dependent variable is binary (the choice of one particular political party), therefore we used the logistic version of the models (Rabe-Hesketh and Skrondal 2008; Snijders and Bosker 1999).

We can evaluate the effect of interviewers and geographic context based on the variance of group-level residuals ( $σ_{settlement}^{2}$ and $σ_{interviewer}^{2}$ ) of the regression models. The comparative strength of the two effects can be judged from their ratio. It is important to note that in multilevel logistic regression models, the size of variances in themselves is to be interpreted with caution, since the individual-level variance is fixed to a more or less arbitrary value for mathematical reasons, which shifts the value of the group-level variance (see Snijders and Bosker 1999:227–29).

We determined the statistical significance of variances, but we aimed to report an effect size, too, which facilitates the interpretation of the substantive significance of interviewer variance and which is directly comparable with the odds ratios measuring the effect of the other (demographic, etc.) variables. We decided to use the median odds ratio (MOR), which was recently suggested by some authors (Larsen et al. 2000; Larsen and Merlo 2005; Ohlsson et al. 2005; Rabe-Hesketh and Skrondal 2008). MOR can be interpreted on the better-known scale of odds ratios rather than relying on the above analysis of variance–type approach; however, to our knowledge, it has been never used before in the analysis of interviewer effects.

Let’s take an example of how to calculate the MOR belonging to the interviewers (MOR_interviewer). Suppose we have two settlements each with three interviewers. The respondents of the three interviewers covering the first settlement had 0.8, 0.6, and 0.5 estimated odds of choosing party X. The odds ratio to go with the pairs made up of the three interviewers was 0.8/0.6 = 1.33, 0.8/0.5 = 1.6, and 0.6/0.5 = 1.2. In the second settlement, party X was less popular: 0.5, 0.4, and 0.3 were the estimated odds for the three interviewers. The odds ratios for the three interviewer pairs were 1.25, 1.33, and 1.66. The median of the six odds ratios is 1.33. It shows that if the respondent is contacted by another interviewer working in the same settlement, the respondent’s odds of choosing party X will (in median) increase 1.33 times. It can also be seen that we have eliminated the effect of the settlement by creating pairs only within the same settlements. Ignoring the settlement factor would have led to a higher MOR for all the pairs created from the set of six interviewers.

Similarly, when calculating MOR_settlement, we compare the party choice odds of respondents living in different settlements but interviewed by the same interviewer. MOR_settlement and MOR_interviewer can be directly compared.

Interviewer effects in multilevel models are, actually, the same as the magnitude of clustering. If beyond measuring the magnitude of clustering we also wish to explain it, we can define the following explanation typology:

Clustering arising from nonrandomized design (technical explanation).

The clustering, in part, can be a result of the differences in composition of respondents across interviewers. This compositional effect can arise as a result of sampling variance or nonresponse variance. We can attempt to separate this components of clustering by including some well-chosen individual-level predictors.

Interviewer-level clustering can also be a result of the model’s inability to completely separate geographic effects from interviewer effects. We can mention two general issues here. One problem causes the overmeasurement of interviewer influence as opposed to settlement influence, while the other vice versa.

If the survey company employs local interviewers, their own political preferences are area determined, leading to real area effects that may look like interviewer effects. However, our data showed no relevant association between the party preferences of interviewers and those of the settlements they covered, so this potential problem doesn’t arise in our case.

Interviewer routines may be locally instructed, leading to real interviewer effects that may look like area effects. We had no data concerning the organization of the interviewers, but as the nontackling of this problem reduces interviewer effects at the expense of larger settlement effects, it will push the interviewer effects measurement toward a more conservative estimation.

Content-based explanation: A partial explanation for interviewer effects can be given by involving some relevant characteristics of the interviewer in the model.

When explaining the political preference of the interviewer (defined as fixed effect and measured with the odds ratio calculated from the regression coefficient), the aim can be to clarify, at least partially, the association–causation relationship, involving potential confounders. This step, however, is often missing from literature on interviewer effect because of the scarcity of information on interviewers. For example, Eisinga et al. (2012) don’t use further interviewer-level variable in the model when analyzing the effect of interviewer body mass index (BMI), yet in the Discussion section, their explanation veers off into the direction of causality without mentioning confounding problems.

When defining the models in our study, we followed the logic above, thus we involved explanatory variables in the empty multilevel model for each level (individual, interviewer, and settlement). The size of MOR can be judged by comparing it to the odds ratios of the explanatory variables.

We defined the dependent binary variable as preference for a given party versus preference for any of the others, leaving uncertain voters (those unable or unwilling to answer) out. The reason for discarding them is that we mainly wanted to find out more about the considerations behind party choice rather than revealing the motives behind steady party preferences. An alternative to the separate multilevel logistic regressions is a single multilevel multinomial logistic regression that contrasts each party with a reference party. For interpretative reasons, we prefer separate binary models contrasting a given party to the rest.

We started with the empty model (model 1), where the only control variable was the month when the interview was conducted. The aim of this model was to measure the two total effects separately.

A range of individual-level predictors (gender, age group, education, economic activity, household structure, and religious faith) were included as fixed effects into model 2 to control for differences in sample compositions (see point 1 in the explanation typology above). We chose variables that, according to the literature, have the strongest influence on interviewer effects and on party preference (Kmetty and Tóth 2011; Tardos 2011). Model 2 also answers the question whether settlement effects may be explained by these variables. Socioeconomic composition of settlements may well characterize the social context in which one’s political decisions are made through direct personal contacts and indirect perceptions (Burbank 1997).

The first stage of the content-based explanation was carried out with the inclusion of the interviewer’s party preference (model 3), then we controlled the findings for further interviewer-level variables (model 4) such as age, gender, income, and education (see points 2 and 3 in the explanation typology above). Models 3 and 4 also answer the question as to which extent settlement effect is explained by settlements’ interviewer composition.

We tried to explain the residual settlement effect by adding unemployment rate and population size of the settlement to the model (model 5) that are considered as relevant characteristics of settlements in current Hungarian political research (e.g., Bálint and Bozsonyi 2012; Kmetty and Tóth 2011; Tardos 2011; Vécsei 2011). Both might play important roles as proxies of economic environment, while the size of the settlement might have decisive impact on the quality and quantity of interpersonal contacts that form our political preferences (Burbank 1997).

We used the xtmelogit command of Stata version 13 to estimate the cross-classified multilevel logistic regression models. It is to be noted here that parameter estimation of a multilevel logistic regression model is complicated compared to a multilevel linear regression model because a numerical approximation is needed. Since the complexity of the model arising from crossed effects extremely increases run time, we used Laplace approximation, which is quicker but yields less accurate results.

Findings

Starting with model 1 (see Table 1), the most noteworthy finding is that judging by the MOR, the interviewer affects the choice of political party at least as strongly as the respondent’s place of residence. Moreover, in the case of MSZP, interviewer effects are around 2, much stronger than that of the settlement, and settlement variance is not statistically significant here. The impact of the settlement is the strongest in the case of the Jobbik, which corresponds with the high geographical heterogeneity the supporters of the party showed in the 2010 parliamentary election.

Table 1.

The Results of Model Building with Statistical Significance Calculated for σ².

Models		Fidesz	MSZP	Jobbik
Model 1 (empty model)	MOR_interviewer	1.50	1.94	1.50
	MOR_settlement	1.48	1.39	1.61
	$σ_{interviewer}^{2}$	0.182*	0.484*	0.179*
	$σ_{settlement}^{2}$	0.167*	0.122	0.249*
	$σ_{interviewer}^{2} / σ_{settlement}^{2}$	1.088	3.988	0.719
	Log likelihood (df = 13)	−3,325.2	−1,936.5	−2,151.7
	AIC	6,676	3,899	4,329
	BIC	6,761	3,984	4,414
Model 2 (respondent’s characteristics)	MOR_interviewer	1.50	2.04	1.55
	MOR_settlement	1.46	1.49	1.64
	$σ_{interviewer}^{2}$	0.181*	0.556*	0.213*
	$σ_{settlement}^{2}$	0.158*	0.174*	0.268*
	$σ_{interviewer}^{2} / σ_{settlement}^{2}$	1.144	3.192	0.794
	Log likelihood (df = 25)	−3,228.8	−1,760.2	−2,084.1
	AIC	6,508	3,570	4,218
	BIC	6,671	3,734	4,381
Model 3 (respondent’s characteristics + interviewer’s party preference)	MOR_interviewer	1.45	1.87	1.51
	MOR_settlement	1.45	1.51	1.64
	$σ_{interviewer}^{2}$	0.154*	0.432*	0.189*
	$σ_{settlement}^{2}$	0.154*	0.188*	0.266*
	$σ_{interviewer}^{2} / σ_{settlement}^{2}$	0.996	2.292	0.710
	Log likelihood (df = 30)	−3,226.4	−1,756.1	−2,082.8
	AIC	6,513	3,572	4,226
	BIC	6,709	3,768	4,421
Model 4 (respondent’s characteristics + interviewer’s party preference + interviewer’s other characteristics)	MOR_interviewer	1.40	1.83	1.40
	MOR_settlement	1.46	1.49	1.69
	$σ_{interviewer}^{2}$	0.124*	0.400*	0.123 ^o
	$σ_{settlement}^{2}$	0.158*	0.176*	0.306*
	$σ_{interviewer}^{2} / σ_{settlement}^{2}$	0.783	2.276	0.402
	Log likelihood (df = 36)	−3,111.9	−1,709.1	−2,013.4
	AIC	6,296	3,490	4,099
	BIC	6,529	3,724	4,332
Model 5 (respondent’s characteristics + interviewer’s party preference + interviewer’s other characteristics + settlement’s characteristics)	MOR_interviewer	1.40	1.82	1.41
	MOR_settlement	1.45	1.48	1.69
	$σ_{interviewer}^{2}$	0.123*	0.391*	0.129
	$σ_{settlement}^{2}$	0.151*	0.167*	0.301*
	$σ_{interviewer}^{2} / σ_{settlement}^{2}$	0.814	2.339	0.427
	Log likelihood (df = 38)	−3,110.5	−1,708.0	−2,013.2
	AIC	6,297	3,492	4,102
	BIC	6,544	3,739	4,349

Note: MOR = median odds ratio; df = degree of freedom; AIC = Akaike information criterion; BIC = Bayesian information criterion, MSZP = Magyar szocialista párt.

*p < .05.

In general, the effect of the interviewer and of the settlement can be said to be significant (the multilevel model fits significantly better to the data than the one-level model, according to the likelihood ratio test, and also, with one exception, both $σ_{settlement}^{2}$ and $σ_{interviewer}^{2}$ are statistically significant).

Having involved individual-level predictors in the model (model 2) significantly improved the fit of the model based on the log likelihood values (Table 1), so they do indeed influence the choice of political parties. Comparing model 2 to model 1 based on the variance ratio, we can say that the comparative strength of the interviewer and settlement effect remained unchanged, with the exception of the MSZP, where the ratio decreased (from 3.988 to 3.192). That is, the effect of interviewers choosing the MSZP can be partially explained by the demographic composition of their respondents.

The impact of individual characteristics on party choice is outside the scope of this article, so they are presented in the Online Appendix. Let us note, however, that both the person of the interviewer and the respondent’s place of residence have at least as strong an impact on their choice as do the demographic characteristics of the respondent. Even educational level and religious faith, which generally are considered important predictors of party preference in Hungary, have, at most, the same impact than the interviewer or the settlement.

For Model 3, arrived at after involving the party preference of the interviewer, the ratio of the interviewer and the settlement variance didn’t show considerable change in comparison with the previous model for Fidesz and Jobbik; yet for the MSZP, the ratio decreased. Thus, the political preferences of the interviewer do have a role in the case of the MSZP. MOR_interviewer for all the three parties still shows a considerable degree of unexplained interviewer effects.

Next, other relevant characteristics of the interviewers were included (Table 1, model 4). Interviewer effects seemed explainable to some extent by these characteristics especially in the case of Jobbik, since the ratio of the interviewer and the settlement variance decreased a lot (from 0.710 to 0.402), and the interviewer variance lost its statistical significance.

For model 5, arrived at after involving settlement-level variables, the ratio of the interviewer and the settlement variance increased only minimally, so the new variables could not capture the impact of geographical context.

Having examined group-level random effects, let us turn our attention to evaluating the explanatory variables on the basis of the estimates of model 5. Interviewer’s party preference has a statistically significant effect on choosing the Fidesz (Table 2), namely, Fidesz-supporter interviewers have the greatest chance to find a Fidesz-supporter respondent. As opposed to them, with an MSZP-supporter interviewer, the odds of choosing Fidesz are only 65%. As for choosing the other parties, the effect of interviewer’s preference is statistically not significant, yet it shows a systematic pattern: The effect of interviewer’s political preference is such that respondents tend to have a preference similar to that of their interviewers.

Table 2.

Regression Coefficients of Model 5 as Odds Ratios.

Variables		Fidesz	MSZP	Jobbik
Interviewer party preference (reference: Fidesz)	MSZP	0.65*	1.84†	1.09
	LMP	0.83	1.16	1.38
	Jobbik	0.73	1.58	1.53
	Other party	0.53†	2.89†	1.44
	Uncertain	0.81	1.36	1.25
Interviewer age-group (reference: 18–29)	30–44 Years	1.40	0.67	0.68
Interviewer age-group (reference: 18–29)	45–59 Years	1.04	0.91	0.83
Interviewer gender: male (reference: female)		1.10	0.77	0.74†
Education of interviewer (reference: vocational)	Secondary nonvocational	0.49	0.46	2.54
Education of interviewer (reference: vocational)	Higher education	0.65	0.53	1.65
Interviewer monthly net income (in 100,000 HUF)		0.96	1.02	0.93
Settlement population size (in 10,000 inhabitants)		0.99	1.02	1.00
Local unemployment rate (per 1,000 inhabitants)		1.00	1.00	1.00

Note: MSZP = Magyar szocialista párt; LMP = Lehet más a politika; HUF = Hungarian forint. †p < .1.

*p < .05.

The data in the table also reveal that an interviewer who supports other parties reduces the probability of the respondent’s picking Fidesz and increases that of picking MSZP (by almost three times) as opposed to a Fidesz-supporter interviewer.

We identified no contrasts that are significant at the 5% level among the demographic characteristics of the interviewers. Settlement characteristics do not have a significant effect either.

Discussion

According to our findings, the interviewer affects party preference. The MOR is around 2 for the MSZP, showing that changing nothing else but the person of the interviewer in half of the cases would result in at least doubling the chance of choosing MSZP. One of our surprising results is that the impact of the interviewer in the majority of the cases is greater, in some cases much greater than that of the geographical context. This finding is in correspondence with others (P. Davis and Scott 1995; O’Muircheartaigh and Campanelli 1998; Schnell and Kreuter 2005). An exception is Turner et al. (2015), who found interviewer effects considerably smaller than the area effect, which, according to them, is likely a result of their dependent variables (travel conditions), which are factual and strongly area dependent.

Interviewer effects can partly be explained by the compositional effect, which would decrease if we controlled for respondent-level characteristics. Our data support the presence of compositional effect in the case of MSZP.

According to our findings, interviewers’ political preference has a significant effect and it does not disappear once controlled for obvious interviewer characteristics. Our findings indicate that to some extent, respondents tend to replicate their interviewer’s party preference. Similar findings were reported for general political attitudes (e.g., Lipps and Lutz 2010), who linked them with the social desirability bias. Our research gives no indication of what channels respondents might use for gauging interviewers’ perceived expectations—further qualitative research should be carried out to shed light on this issue.

Interviewer effects seemed explainable to some extent by demographic characteristics of the interviewer in the case of Jobbik. When explaining this finding, it must be noted that according to researchers looking at Hungary in 2010, many citizens preferred to hide their preference for the far-right extremist party Jobbik. Suppose that both Jobbik-supporter interviewers and respondents tried to avoid expressing their preference openly, interviewer demographic characteristics could have a greater role in establishing respondent’s trust to reveal his or her party preference.

The importance of interviewer effects is demonstrated by the fact it had at least as great an effect on party preference than do the demographic characteristics of the respondent. All this may indicate that party preference is more of a decision shaped in the course of social interactions and may even be subject to change, here by the interviewer. This finding contradicts those political approaches that assume that political behavior is a determinism predictable by sociodemographic characteristics.

We found different effect sizes across the different parties, with that of the MSZP being the largest. Apparently, admitting a preference for a party that had a disappointing performance while in power and then suffered defeat was perceived as a kind of pledge of allegiance. As a result, some of the MSZP voters might have decided not to disclose their preference or pretended to support another party. But some interviewers might reduce this hiding behavior by relying on their experience and attitudes (as the interviewer party preference effect in Table 2 might show). The Hungarian Gallup Institute actually supplied some evidence for people’s tendency to hide their MSZP preference (“Az MSZP és a Jobbik fej-fej mellett küzdhet a második helyért (Gallup) [MSZP and Jobbik abreast in race for second place]” 2010).

Overall, our research highlights the importance of interviewer effects for face-to-face survey data collection. As for the practical implications: (1) The findings highlight the importance of standardizing interviewer behavior. Quantifying interviewer effects can prove useful when comparing different survey modes, companies, or questionnaire items. The fact that interviewer effects exceed the impact of geographical context, in turn, leads us to the theoretical conclusion that: (2) When examining the impact of geographical context on a social phenomenon, ignoring interviewer effects will result in a strong overestimation of the geographical effect. Also, (3) apart from the other characteristics of sampling (e.g., clusters, stratification), the interviewer is an important part of the survey design—thus, a mathematically correct analysis would be one that incorporates the interviewer identifier over and above the primary sampling unit.

Supplemental Material

Supplemental Material, nemeth_FM_8-10-16_supplement_final - Strong Impact of Interviewers on Respondents’ Political Choice: Evidence from Hungary

Supplemental Material, nemeth_FM_8-10-16_supplement_final for Strong Impact of Interviewers on Respondents’ Political Choice: Evidence from Hungary by Renáta Németh and Alexandra Luksander in Field Methods

Footnotes

Acknowledgment

We acknowledge the helpful comments of the anonymous reviewers, one of whom suggested to add point 1b to the Method section.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Renáta Németh’s work was supported by the János Bolyai Research Scholarship from the Hungarian Academy of Sciences and by the Faculty of Social Sciences, Eötvös Loránd University, Hungary.

Supplemental Material

Supplementary material for this article is available online.

Note

References

Az MSZP és a Jobbik fej-fej mellett küzdhet a második helyért (Gallup) [The MSZP and the Jobbik are fighting neck-and-neck for the second position]. 2010. https://www.portfolio.hu/gazdasag/az-mszp-es-a-jobbik-fej-fej-mellett-kuzdhet-a-masodik-helyert--gallup.130682.html (accessed February 20, 2017).

Bálint

Bozsonyi

. 2012. Választói részvétel és véleménypolarizáció térfilterezett modelljei [Spatially filtered models of voter participation and opinion polarization]. In Változó képletek, változatos perspektívák. Tanulmánykötet Tardos Róbert 65. születésnapjára, eds. Kmetty

Koltai

, 251–74. Budapest: Háttér Kiadó.

Berinsky

A. J.

1999. The two faces of public opinion. American Journal of Political Science 43:1209–30.

Biemer

Stokes

. 1985. Optimal design of interviewer variance experiments in complex surveys. Journal of the American Statistical Association 80:158–66.

Burbank

M. J.

1997. Explaining contextual effects on vote choice. Political Behavior 19:113–32.

Davis

D. W.

Silver

B. D.

. 2003. Stereotype threat and race of interviewer effects in a survey on political knowledge. American Journal of Political Science 47:33–45.

Davis

Scott

. 1995. The effect of interviewer variance on domain comparisons. Survey Methodology 21:99–106.

Eisinga

Grotenhuis

M. T.

Larsen

J. K.

Pelzer

. 2012. Interviewer BMI effects on under- and over-reporting of restrained eating: Evidence from a national Dutch face-to-face survey and a postal follow-up. International Journal of Public Health 57:643–47.

Fowler

F. J.

Mangione

T. W.

. 1990. Standardized survey interviewing: Minimizing interviewer-related error. Beverly Hills, CA: Sage.

10.

Hox

J. J.

1994. Hierarchical regression models for interviewer and respondent effects. Sociological Methods and Research 22:300–18.

11.

Hyman

H. H.

1954. Interviewing in social research. Chicago: University of Chicago Press.

12.

Katz

1942. Do interviewers bias polls? Public Opinion Quarterly 6:248–68.

13.

Kish

1995. Methods for design effects. Journal of Official Statistics 11:55–77.

14.

Kmetty

Tóth

. 2011. A politikai részvétel három szintje (The three levels of political participation). In Részvétel, képviselet, politikai változás, eds. Tardos

Enyedi

Szabó

, 75–115. Budapest: Demokrácia Kutatások Magyar Központja Alapítvány.

15.

Krysan

Couper

M. P.

. 2003. Race in the live and the virtual interview: Racial deference, social desirability, and activation effects in attitude surveys. Social Psychology Quarterly 66:364–83.

16.

Lansford

, ed. 2014. Political handbook of the world 2014. Thousand Oaks, CA: Sage.

17.

Larsen

Holm

Budtz-Jørgensen

Endahl

. 2000. Interpreting parameters in the logistic regression model with random effects. Biometrics 56:909–14.

18.

Larsen

Merlo

. 2005. Appropriate assessment of neighborhood effects on individual health—Integrating random and fixed effects in multilevel logistic regression. American Journal of Epidemiology 16:81–88.

19.

Lipps

Lutz

. 2010. How answers on political attitudes are shaped by interviewers: Evidence from a panel survey. Swiss Journal of Sociology 36:345–58.

20.

Nishizawa

Kuriyama

. 2016. Social desirability bias and the Japanese voters. In Electoral survey methodology: Insight from Japan on using computer-assisted personal interviews, eds. Kohno

Nishizawa

, 57–73. London: Routledge.

21.

Ohlsson

Lindblad

Lithman

Ericsson

Gerdtham

U. G.

Melander

Råstam

Merlo

. 2005. Understanding adherence to official guidelines on statin prescribing in primary health care—A multi-level methodological approach. European Journal of Clinical Pharmacology 61:657–65.

22.

O’Muircheartaigh

Campanelli

. 1998. The relative impact of interviewer effects and sample design effects on survey precision. Journal of the Royal Statistical Society 161:63–77.

23.

Pickery

Loosveldt

. 2002. A multilevel multinomial analysis of interviewer effects on various components of unit nonresponse. Quality and Quantity 36:427–37.

24.

Rabe-Hesketh

Skrondal

. 2008. Multilevel and longitudinal modeling using Stata. College Station, TX: State Press.

25.

Schnell

Kreuter

. 2005. Separating interviewer and sampling-point effects. Journal of Official Statistics 21:389–410.

26.

Shoemaker

P. J.

Breen

Stamper

. 2000. Fear of social isolation: Testing an assumption from the spiral of silence. Irish Communications Review 8:65–78.

27.

Snijders

T. A. B.

Bosker

. 1999. Multilevel analysis: An introduction to basic and advanced multilevel modelling. London: Sage.

28.

Tardos

2011. Konszolidált szavazóközönség eróziós tendenciákkal. A 2010-es választások a részvételi aktivitás és szavazói motívumok fényében [A consolidated voting community with tendencies of erosion. The 2010 elections in Hungary in the light of participation and voters’ motives]. In Új képlet. Választások Magyarországon, 2010, eds. Enyedi

Szabó

Tardos

, 273–374. Budapest: Demokrácia Kutatások Magyar Központja Alapítvány.

29.

Turner

Sturgis

Martin

Skinner

. 2015. Can interviewer personality, attitudes and experience explain the design effect in face-to-face surveys? In Improving survey methods: Lessons from recent research (European Association of Methodology Series), eds. Engel

Jann

Lynn

Scherpenzeel

Sturgis

, 72–75. Abingdon, UK: Routledge.

30.

Vécsei

2011. A 2006 és 2010 közötti területi politikai szerkezetváltozás és a területi reálfolyamatok összefüggéseinek alakulása [Tendencies in the correlations between the territorial real processes and the structural changes in the political field in the 2006–2010 period in Hungary]. In Részvétel, képviselet, politikai változás, eds. Tardos

Enyedi

Szabó

, 383–407. Budapest: Demokrácia Kutatások Magyar Központja Alapítvány.

31.

West

B. T.

Blom

A. G.

. 2016. Explaining interviewer effects: A research synthesis. Journal of Survey Statistics and Methodology 5:175–211.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB