So far, it is apparent that secondary survey research is typically based on multiple surveys conducted over relatively long periods of time. This suggests that secondary research encounters problems quite different from primary research. What are they? Here we look at secondary research based on multiple surveys (at least two, and averaging about five), and we broaden our sampling procedure to include fifty-four examples of journal articles, recent papers at major academic conferences, and the latest editions of published textbooks in public opinion. We find three important issues that need to be considered by secondary researchers in their handling of survey data:

•  Did the research compare poll questions that originally had different wording? As is well known, even modest and seemingly innocuous wording changes can significantly change survey responses. Over time, survey customs on wording have varied as to whether they allow a middle response option or a "don't know" or "no opinion" option. At one time it was common to ask respondents simply to agree or disagree with statements; later, the "agree or disagree" or "or not" or "or do you disagree" balanced format became common. Practices also vary among survey houses. As Figure 5 indicates, nearly half (44 percent) of the polls cited had some form of wording switch, most often a variation of "don't know" or "no opinion," but also including other wording changes, such as a switch from "Negro" to "black."

•  Did surveys use similar contacting methods over time? At one time most surveys were conducted face-to-face at the respondent's door. By the 1970s telephone contacts became more widespread for reasons of cost, timeliness, oversight, and (arguably) callbacks. Other contacting strategies, such as email and website panels, remain less common, at least in these journals, but they may become more prevalent in the future. Figure 5 indicates that interview mode switches happen fairly often. Again, 44 percent of published secondary survey research includes such a switch.

•  Does secondary survey research often switch polling houses? This possibility most frequently arises when one polling house borrows questions originally used by another. A researcher may be inclined to include all the questions available even when they come from different polling houses, since this increases the number of data points available. Compounding the problem, nonacademic survey researchers often do not report their interviewing procedures, such as within-house respondent selection procedure, whether substitutions are allowed, or the number of callbacks. House effects may affect results in ways that are difficult to determine for even the most industrious secondary survey researcher, and probably impossible for a reader to determine. As Figure 5 shows, polling house switches occur in 6 percent of the published secondary survey research sampled here.