Abstract
In 1066, when William the Conqueror’s fleet sailed from Normandy toward the English coast, only two of the 700 ships were lost at sea. However, one of those two contained the expedition’s soothsayer. “It’s no great loss,” said William, “he couldn’t even predict his own fate.”
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
In earlier writing, I and most others have often used “context effects” and “question-order effects” more or less interchangeably. However, Bishop, Hippler, Schwarz, and Strack (1988) have pointed up one important difference by showing that context effects can occur in self-administered questionnaires where question order is not controlled by the investigator. In addition, some order effects (e.g., due to “fatigue”) may not be a matter of context in the sense of “transfers of meaning” (Schuman & Presser, 1981, p. 23). These are important theoretical distinctions to keep in mind, although in this chapter I shall not attempt to be so precise in language at every point.
Several reports presented no numbers at all, which gave them about the same scientific status as accounts of individual sightings of a monster in Loch Ness. Other reports (e.g., Duverger, 1964) gave percentages supposedly due to context but provided neither Ns nor significance tests that allowed evaluation of the null hypothesis.
T. W. Smith (1988a) later carried out a somewhat similar analysis for the General Social Survey and drew the same conclusion.
Confidence intervals can be calculated based on the pooled data from the nine experiments: 6.2% to 15.2% for the difference between two means, based on the assumption of simple random sampling. Four of the differences in Table 2.2 fall outside this interval, but only in one case (July 1983) is the deviation very large.
In addition to the nine replications on national samples shown in Table 2.2, I repeated the experiment as part of a 1990 face-to-face survey with University of Michigan undergraduates. The right to an abortion was supported for the married woman by 79% of those receiving the question first (N = 251) but by only 62% of those receiving the same question after answering the question about the defective fetus (N = 243); chi-square= 16.8, df= 1, p <.001. Thus the effect is, if anything, even stronger in a population of sophisticated young adults whose overall level of support for legalized abortion is higher than is the case for the general population.
Statisticians like to point out that with large enough samples the null hypothesis will almost always be rejected. However, with the size samples typically used in survey research, this is definitely not the case, and furthermore any practical increase (e.g., doubling) of sample size is much more likely in my experience to eliminate a “borderline” effect taken too seriously than it is to discover a trivial real effect.
To keep the present discussion brief, only the effect on the Communist reporters item is shown and discussed here, but the American reporters item also shows the effect of the reciprocity norm, although in the reverse direction (responses to the American reporters item decrease following the Communist reporters item). Moreover, both effects are located primarily among respondents who agree with the American reporters item when it comes first, or disagree with the Communist reporters item when it comes first-producing what T. W. Smith (1982) calls “conditional order effects.” See Schuman and Presser (1981, pp. 28–31) for the full results with the two items, including the interaction with time discussed below.
Valuable further work on the basic effect involving self-reports of political interest is summarized by Bishop (1987).
One safeguard that cannot be counted on is the traditional assumption of survey researchers that an item is not likely to be affected by an earlier item if the two are in separate parts of the questionnaire. See Bishop (1987) and Schuman, Kalton, and Ludwig (1983) for evidence against this assumption. See also Ottati, Riggle, Schwarz, and Kuklinski (1989) and Schwarz et al. (1991) for further complexities regarding number and spacing of items.
The percentage of the population agreeing with abortion in the case of rape is almost exactly the same as in the case of a possible defect in the baby, and the percentage agreeing in the case of the health of the mother is even higher. In addition, although it has been possible to substitute for the married woman item another GSS item on a woman wanting an abortion because she is poor (which yields about the same agreement) and obtain a significant effect in combination with the defect item, the rape/low-income combination also failed to produce a significant context effect (p >.10).
A more complete account through 1988 of our experiments on the abortion context effect can be found in Scott (1988).
When a question on limiting Japanese products from entering the United States was asked alone and those respondents who opposed such limitations were probed, most of them cited a concern over reciprocity as their reason. The use of context to make the norm salient to all simply increases the proportion answering on that basis (see Schuman & Ludwig, 1983, p. 117).
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 1992 Springer-Verlag New York, Inc.
About this chapter
Cite this chapter
Schuman, H. (1992). Context Effects: State of the Past/State of the Art. 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_2
Download citation
DOI: https://doi.org/10.1007/978-1-4612-2848-6_2
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