Full text: Brückner, Hannah: Surveys don't lie, people do?

-60- 
Individuals matter. Only four interviewer dummy variables are significant in this model but it 
is noteworthy that compared to the five interviewers who completed under 30 cases they are all 
positive and large. Adding the interviewer dummy variables improves the model fit (LR-79.8 with 19 
df). Three of these interviewers have also positive and significant effects on visible errors. When 
choosing the interviewer with the highest work load as comparison category, four interviewers emerge 
with large significant and negative coefficients. This finding points to the importance of improving 
field control; neither additional training, nor feed-back from the editing group nor even supervision 
during the interviewing helped to prevent a situation in which some of the most active interviewers 
in the field produced consistently below-standard data quality, at least in terms of the two measures 
employed in this analysis. 
4.4. How Much Do Individuals Matter? 
A comparison of the results from descriptive and multivariate analysis in regard to the 
difference between the two types of error leads to the question whether the different effects of 
experience and productivity on the two types of error hold for all interviewers alike. Therefore, I 
introduced interaction terms with interviewers and both experience and productivity. To avoid 
problems with collinearity and convergence, I introduced interaction terms in groups of four or five." 
In the model for visible errors, interactions of interviewers with productivity yielded no better model 
fit except for two interviewers (No. 15 and 16 in Table V), for whom the effects were positive and 
significant. For the model of invisible errors, on the other hand, where the effect of productivity was 
negative, for two interviewers the effect was positive and significant (No. 6 and 13). I found only one 
significant interaction effect of interviewers and experience on invisible errors which was positive, in 
contrast to the overall effect, which is expected to be negative. For three interviewers the effect of 
experience on visible errors was negative and significant, for another three positive. 
When using the predicted main effect of individuals or overall workload as classification 
criteria, a pattern giving meaning to these interaction effects did not emerge (e.g. that the effects of 
experience and productivity were systematically different for good' interviewers or for bad' interview- 
„Due to limitations of LIMDEP, I could not introduce all interaction terms at once in the 
model. When doing so in the L-OLS model, the improvement in fit for visible errors is not 
significant although three interaction terms are significantly different from zero (one with a 
positive sign, two with a negative sign). In the L-OLS model for invisible errors, multi-collinearity 
precludes further analysis.
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.

powered by Goobi viewer