Abstract
It has been widely recognized that in personal interviews the response to a specific survey question can be seriously affected by the context of that question, that is, the preceding parts of the questionnaire (Cantril, 1944; Schuman & Presser, 1981, pp. 23–56). More specifically, responses can be affected by a variety of factors: prior questions and their answers, response scales of preceding questions, introductions to questions, and tasks that the respondents have completed (Tourangeau & Rasinski, 1988). Context effects are response effects coming from one or more preceding questions (and answers) or from response scales belonging to previous questions.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Notes
The term “context effects” is used here in a broader sense than in Schuman and Presser (1981, p. 23), where it refers to question-order effects involving transfers of meaning. For other more mechanical types of artifacts, they use the term “sequence effects.”
Since voting is compulsory in Belgium, we were able to use the electoral registers (per municipality) of the general elections (October 1985), which are exhaustive lists of all adults. These registers include information on age, sex, marital status, occupation, and address of each resident of voting age. A random sample of 1,000 married women between the ages of 21 and 55 was drawn, and 400 sample units were randomly assigned to one of two experimental conditions (Form A and Form B). In view of the replacement for nonresponse, the other units were matched with the original units using the information about occupation and age. With respect to the replacement, special attention was paid to the comparability of the two experimental conditions. The fieldwork was carried out by the market research institute DIMARSO/ GALLUP BELGIUM. All of the interviews were tape recorded.
None of the differences in response distributions has a probability less than.20 under HO.
We found 23 effects that were significant at the.05 level; 4 others exceeded the.10 significance level.
In principle before a minister of one of the official religious denominations but in fact nearly always before a Roman Catholic priest.
The variable “church involvement” was operationalized by three questions: one about baptism, one about the respondent’s actual (subjective) religious or philosophical convictions, and one about church attendance. The categories of the variable are “not catholic,” “nonpracticing Catholic,” “irregular churchgoer,” and “regular churchgoer.” Because of the small sample size, for the analysis of multiway tables these four categories are collapsed into two. The irregular and the regular churchgoers are the “practicing Catholics.”
We can choose between log-linear modeling, logistic regression, and analysis of variance. Logistic regression uses a logit model based on the cumulative logistic probability function (Pindyck & Rubinfeld, 1981). It is very similar to log-linear analysis with a response variable. For both analysis of variance and logistic regression, we need to code the response variable (legal marriage, cohabitation) into two categories: approval (or disapproval) or not. Because this results in a loss of information (“indifferent”), log-linear modeling is preferred.
In factor-response models, only the associations and interactions in which the response variable is included are considered. The associations and interactions between the factors (AOI) need no test and are included in all of the models. In this way, the 167 possible hierarchical models can be reduced to only 20. Since the models are hierarchical, the three-way interaction term AOI includes the terms A, I, O, AI, AO, OI, and AOI.
AIC is based on the degree of correspondence between a selected model and the unknown “true” model that is measured by Kullback-Leibler’s Information measure (the so-called “negative entropy”). AIC is a function of the number of observations in the multiway table (N), the cell frequencies, the number of cells, the likelihood ratio (L2), and the degrees of freedom (df). Within the context of a particular table, AIC depends only on L2 and df, since the other parameters are fixed. In log-linear models, AIC corrects the log likelihood ratio for the number of degrees of freedom (AIC L2 — 2df). Therefore, it may be considered a formalization of the idea of “sparseness.”
We use the computer program LOG-SCAN, developed by L. Daemen: A program to select log-linear models by minimizing Akaike’s Information Criterion, Version 1.0, Department of Sociology, K.U. Leuven.
Form A: “You have had your child(ren) baptized. What were your main reasons for having the child(ren) baptized?” (Insist: Is there any other reason?) Form B: “You have had your child(ren) baptized. On this card (hand card 3) there are a number of reasons why parents have their children baptized. We are going to read them together and afterwards I will ask you for the three most important reasons why you had your child(ren) baptized.” (Some of the reasons are the following: part of religious community [1]; baptism enables to make … a church wedding [4]; to have a religious upbringing [5]; baptism is a sacrament conferring the grace of God [10].)
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 1992 Springer-Verlag New York, Inc.
About this chapter
Cite this chapter
Billiet, J.B., Waterplas, L., Loosveldt, G. (1992). Context Effects as Substantive Data in Social Surveys. In: Schwarz, N., Sudman, S. (eds) Context Effects in Social and Psychological Research. Springer, New York, NY. https://doi.org/10.1007/978-1-4612-2848-6_10
Download citation
DOI: https://doi.org/10.1007/978-1-4612-2848-6_10
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4612-7695-1
Online ISBN: 978-1-4612-2848-6
eBook Packages: Springer Book Archive