Notebook, Uncategorized

“Who says what” in multiple choice questions. A comprehensive exploratory analysis protocol

By M. Landaluce-Calvo, Ignacio García-Lautre, Vidal Díaz de Rada, & Elena Abascal

The aim of much sociological research is to assess public opinion, and the data are often collected by the survey method. This enables the detection of different response, behaviour or opinion profiles and the characterization of groups of respondents with similar views on a certain topic or set of questions. As is widely known, however, different types of question not only yield different qualities of response, but also require different methods of analysis.

Any attempt to classify survey question types require consideration of five criteria: 1) degree of freedom in the response; 2) type of content, 3) level of sensitivity/threat; 4) level of measurement; and 5) number of response options per question. The last classification (with respect to the number of responses) first differentiates between single response and multiple response questions. Here is the main objective of our article in IJSRM: How to extract maximum information from multiple response questions.

There are two broad types of multiple-response questions. One is the categorical response question, where the respondent is instructed to “check-all-that-apply” (the categories are exhaustive, but not mutually exclusive.). The other is the binary response question, where the respondent is required to check yes or no to each response option. Respondents find “check-all-that-apply” questions more difficult to answer because the multiple options require more use of memory. Under the binary-response format the respondent must consider pairs of options, one by one, and check one option in each case. Each pair of options requires an answer, so only a minimal demand is placed on memory. This procedure yields more responses, in both telephone and online surveys and requires less effort on the part of the respondent, although it may lengthen the questionnaire.

Those admitting various response options can be further classified into grid or check-all-that-apply questions. In the case of the latter, the categories are exhaustive, but not mutually exclusive. This multiple-response question format is its widespread use both in the field of opinion polling and in sociological and marketing research. International research project such as the European Social Survey and the Word Values Survey, for example, contain large numbers of multiple responses questions.

All the above considerations relate to the stages of data collection and participant opinion retrieval, but what about the analysis? A review of the specialist literature reveals a lack of attention to the specific data-processing treatment, and the failure to use a multidimensional exploratory approach that would enable the maximum amount of information to be extracted from the response options. The analysis is limited mainly to calculating one-dimensional frequencies (the frequency with which a given response occurs over the total number of respondents or total number of responses) or two-dimensional frequencies resulting from crossing the chosen response option with other socio-demographic or socio-economic characteristics, etc; in other words, a partial approach in either case.

Our article in IJSRM present a multidimensional analysis protocol that provides the researcher with tools to identify more and better profiles about “who says what”. The underlying philosophy in this approach is to “let the data speak for themselves”, and to learn from them. The strategy begins by coding the response options as a set of metric binary variables (presence/absence). The ideal methodological duo for the exploration of the resulting data is Principal Component Analysis coupled with an Ascending Hierarchical Cluster Analysis, incorporating, in addition, supplementary variables (gender, age, marital status, educational attainment, etc.).

This protocol applies to the analysis of three different multiple-response questions included in a Spanish National Sociological Survey (CIS- Centro de Investigaciones Sociológicas):

  1. “How do you usually spend your free time?”, the respondent has 17 options and can select as many as desired; no order of preference is required and the categories are not mutually exclusive.
  2. “During 2017, how have you spent or do you intend spending your leisure periods?”, with 10 options, there is no limit on the number of them that can be checked, but there are two which automatically exclude the rest: “I haven’t thought about it yet” and “I have no leisure periods”.
  3. When deciding how to spend your days off, what are your top three priorities?”, there is alimit of three options, out of 10 possible, no order of preference required.

This empirical analysis provides evidence not only of the interpretation potential of the coding/analysis protocol, but also of the limitations of some multiple-response question formats. Specifically, it is shown that multi-response with limited options is not a suitable format for detecting response patterns or overall tendencies leading to the identification of global respondent profiles. In addition, this study corroborates that in the “forced choice” and “check all that apply” the respondents are more likely to choose from the options presented at the beginning of a list (primacy effect). Early theories attributing the phenomenon to such questions requiring deeper cognitive processing.

Read the full article in IJSRM here.

featured, Notebook

I Say, They Say: Effects of Providing Examples in a Survey Question

By Eva Aizpurua, Ki H. Park, E. O. Heiden & Mary E. Losch

One of the first things that survey researchers learn is that questionnaire design decisions are anything but trivial. The order of the questions, the number of response options, and the labels used to describe them can all influence survey responses. In this Research Note, we turn our attention to the use of examples, a common component of survey questions. Examples are intended to help respondents, providing them with information about the type of answers expected and reminding them of responses that might otherwise go unnoticed. For instance, the 2020 U.S. National Health Interview Survey asked about the use of over-the-counter medication, and included “aspirin, Tylenol, Advil, or Aleve” in the question stem. There are many other examples in both national and international surveys. Despite the potential benefits of using examples, there is a risk that respondents will focus too much on them, at the expense of overlooking cases not listed as examples. This phenomenon, called the “focusing hypothesis”, is what we test in our study.

Using an experimental design, we examined the effects of providing examples in a question about multitasking (“During the time we have been on the phone, in what other activities, if any, were you engaged [random group statement here]?”). In this experiment, respondents were randomly assigned to one of three conditions: the first group received one set of examples (watching TV or watching kids), the second group received a different set of examples (walking or talking with someone else), while the final group received no examples. Our goal was to determine whether respondents were more likely to report an activity (e.g., watching TV or walking) when it was listed as an example. We also wanted to understand whether providing examples resulted in respondents listing more activities beyond the examples.

We embedded this experiment in a telephone survey conducted in a Midwestern U.S. state and found support for the focusing hypothesis. As anticipated, respondents were more likely to mention the activity if it was provided to them as an example. However, the effect sizes were generally small and examples did not have an effect on the percentage of respondents who identified themselves as multitaskers, nor on the number of activities reported by them. This is because people faced with the experimental conditions were more likely to list the examples presented to them (i.e., watching TV, watching kids, walking, talking with someone else), while those in the control group more frequently reported activities outside this range (cooking, doing housework…), yielding no differences on the frequency of multitasking or on the number of multitasking activities.  Although examples can help respondents understand the scope of the question and remind them of certain responses, the results from this study indicate that they can also restrict the memory search to the examples provided. This has implications for survey practice, suggesting that the inclusion of examples in questions should be carefully considered and limited to certain situations, such as questions in which recall errors are anticipated or when the scope of the question might be unclear.

To learn more, see full IJSRM article here.