For many inquiries in the field of communication research, the analysis of change is of great value. Classic diffusion research in the communication sciences deals with the diffusion of information in society, and it is important to know the dynamics of the diffusion process, as well as the factors that influence it. Media research considers media effects mostly as individual changes caused by media content. Therefore, methods that allow the analysis of these changes are of particular importance in communication research.
There is a fundamental distinction between cross-sectional studies and longitudinal studies. In the case of cross-sectional studies, a survey is conducted only once. Longitudinal studies, on the other hand, are those studies in which the same survey is conducted several times at certain intervals. Only this second type of study allows for a dynamic approach, a longitudinal analysis. There are two distinct types of longitudinal analysis: trend studies and panel studies. The difference between these two types of surveys is that a trend study employs different respondents with each survey, while a panel study implies the use of the same respondents each time for a multiple survey. In other words, a trend study employs a new sample for each measuring time, while a panel study employs the same sample over a series of measuring times. The two different types of longitudinal studies have a variety of advantages and disadvantages, and accordingly call for, as well as permit, different analytical techniques.
Trend studies can be conducted in relatively quick succession. This helps establish with relative accuracy the point in time when changes of specific variables occur. It is also possible to conduct an analysis according to sub-groups, and to examine in which social groups the development shows a greater dynamic. A larger number of data measured at regularly spaced intervals allows for the application of time-series analysis.
However, trend studies have certain limitations. Trend studies draw new independent samples at different points in time, which, nevertheless, have to come from the same population. This does not permit the study of changes at an individual level, but only at an aggregate level. It is therefore impossible to assess which people have changed their opinion over time. Moreover, there may be differences between the samples drawn at different survey times that have nothing to do with changes in the population, but rather stem from time-related variations in the effect of the sampling error. Fluctuations between different survey times may therefore be coincidental. Pertinent statistical analysis may, however, help establish the likelihood of these coincidental differences, and the degree of certainty with which one may interpret the differences found.
Panel studies have traditionally played an important role in communication research. This probably has to do in part with the fact that some of the pioneering communication studies are panel studies, the first among them being The people’s choice (Lazarsfeld et al. 1948, 1st pub. 1944). Their importance, however, is due particularly to the considerable strategic advantages this survey method offers for communication research (Lazarsfeld & Fiske 1938; Lazarsfeld 1940).
The repeated questioning of the same respondents allows for the collection of more data about them. It also permits the building of trust between surveyor and respondent, which in turn may make the posing of more intimate questions possible. There is greater reliability in the assessment of changes, since in the comparison of two independent samples the selection error, which can have a different effect on each of the samples, always has to be taken into account. Panels are an especially useful method to help answer central questions in the communication sciences. Measuring data at certain intervals facilitates the analysis of changes at the individual level as well as the statistical analysis of the conditions under which such changes are most or least likely to occur. This is particularly important for media effects research.
Advantages Of Panel Studies
Panel studies have an important analytical advantage in that they can uncover gross changes. As a rule, social situations are subject to opposing forces. The actual dynamic of social changes can therefore be grossly underestimated. A marketing campaign, for example, may lead to a better assessment of the product in question by a sector of the population while causing negative reactions in another sector. If in this case we compare two independent samples, we will underestimate the actual change caused by the marketing campaign. Only a panel that measures individual changes will provide a clear picture. The example given here shows this clearly (Tables 1 and 2). Table 1 shows the net change in a similar way to a trend study. The picture is one of relative stability. If we consider the sum of all changes, the net change amounts to no more than 5 percent.
An analysis based on individual changes as provided by a panel study provides a more dynamic picture (Table 2). The diagonal thereby represents stability, i.e., here we find the respondents whose position has not changed. This shows that only 62 percent of respondents did not change their opinion. The respondents in the other categories have all changed their opinion. This shows clearly that substantial fluctuations can occur in a seemingly very stable situation. Only in the sum do these fluctuations cancel each other out almost completely. These analyses stress the ways in which panel studies can improve the description of social dynamics.
Panels are generally considered to be a useful method for the analysis of causal relationships. In this context we have to distinguish between two different types of questions. On the one hand, there are the quasi-experimental inquiries. These are analyses that evaluate hypotheses about the influence of one variable on another. Typical questions in media research, for instance, are concerned with assumptions regarding the influence of television on particular cognitive patterns. In this case the analysis follows experimental research models, in that one differentiates two groups, one of which has, for example, watched relevant television programs while the other one has not. The research focus here is on potential differences in the development of these two groups. This type of process is of course only quasi-experimental, since first of all these groups are self-selected and, moreover, the researcher has not manipulated the experimental stimulus.
Table 1 Analysis of trends in attitudes toward the census Question: “In general: are you in favor of or against the census?”
Table 2 Analysis of the actual change in the panel Question: “In general: are you in favor of or against the census?”
On the other hand, the panel study is useful in deciding between conflicting hypotheses about the causal relationship between two correlated variables. In this context the panel study essentially follows the logic that causal relationships can be transferred over into time-related ones, and that cause precedes effect. This means that from the occurrence of the cause, predictions can be made regarding the future effect, while occurrence of the effect does not allow a prognosis regarding a future occurrence of the cause. Thus, if the condition of the variable Y at the point in time t + 1 can be more easily predicted from the condition of the variable X at the point in time t than the other way around, then X is considered the cause and Y the effect. Pertinent analytical procedures based on this logic are, for example, Lazarsfeld’s 16-fold table (1972), Coleman’s calculation of transition probabilities (1964), and the cross-lagged panel-correlation analysis (CLPC; see Campbell 1963; Campbell & Stanley 1963). This last one can be considered to be the standard method in causal analysis in panel studies (McCullough 1978).
CLPC analysis compares the time-lagged correlations of the variables. Its purpose is to find out whether the relationship between variable X at time t and variable Y at time t +1 is stronger than the relationship between Y(t) and X(t + 1). If it is true that X(t)Y(t + 1) > Y(t)X(t + 1), then X is considered to be the causative variable. However, it is important to calculate only cross-correlations that have been adjusted for the autocorrelation of the respective dependent variable. Otherwise the stationary correlation at time t influences the result.
Problems With Panel Studies
A problem with panel studies is that the method requires a greater willingness to cooperate on the part of the respondents. They have to be available for repeated questioning and give their permission for their private address information to be collected. It is often the case that respondents do not take part in all surveys, so that the sample size progressively decreases. This phenomenon is called panel mortality. Panel mortality presents a problem for two reasons. On the one hand, the first sample has to be substantial to ensure it is still large enough at the end of the study. On the other hand, panel mortality presents a serious problem if dropouts are not coincidental but systematic. In such cases panel mortality distorts the sample results.
The terms “panel effect” and “conditioning effect” describe the phenomenon that respondents change their answers due to repeated questioning in the panel. We distinguish two types of panel effects: those in which respondents actually change their position and those in which they merely change their answering behavior. In the first case we have to assume that respondents may have become more critically aware or sensitive regarding the topic of the survey or that they may have become more stable in their opinion. There are very few studies concerning these effects. As a rule, these “conditioning effects” are considered to be very small (Holt 1989). In the second case, it is assumed that the quality of the answers has improved (Waterton & Lievesley 1989). The respondents learn the requirements of the survey interview procedure and build up trust in the surveyors. This can be seen on the whole as a positive effect, but it can also have negative consequences. If the quality of answers improves, this also means that change cannot be assessed with precision. If change occurs we do not know whether it is due to an actual change or merely to the fact that answers have become more precise.
- Campbell, D. T. (1963). From description to experimentation: Interpreting trends as quasiexperiments. In C. W. Harris (ed.), Problems in measuring change. Madison. WI: University of Wisconsin Press, pp. 212–242.
- Campbell, D. T., & Stanley, J. C. (1963). Experimental and quasi-experimental designs for research. Chicago, IL: Rand McNally.
- Coleman, J. S. (1964). Introduction to mathematical sociology. New York: Free Press.
- Holt, D. (1989). Panel conditioning: Discussion. In D. Kasprzyk, G. J. Duncan, G. Kalton, & M. P. Singh (eds.), Panel surveys. New York: Wiley, pp. 340–347.
- Lazarsfeld, P. F. (1940). “Panel” studies. Public Opinion Quarterly, 4, 122–128.
- Lazarsfeld, P. F. (1972). Mutual relations over time of two attributes: Review and integration of various approaches. In M. Hammer, K. Salzinger, & S. Sutton (eds.), Psychopathology: Contributions from the Social, Behavioral, and Biological Sciences. New York: Wiley, pp. 461– 480.
- Lazarsfeld, P. F., & Fiske, M. (1938). The “panel” as a new tool for measuring opinion. Public Opinion Quarterly, 2, 596 – 612.
- Lazarsfeld, P. F., Berelson, B., & Gaudet, H. (1948). The people’s choice: How the voter makes up his mind in a presidential campaign. New York: Columbia University Press. (Original work published 1944).
- McCullough, C. B. (1978). Effect of variables using panel data: A review of techniques. Public Opinion Quarterly, 42, 199–220.
- Waterton, J., & Lievesley, D. (1989). Evidence of conditioning effects in the British social attitudes panel survey. In D. Kasprzyk, G. J. Duncan, G. Kalton, & M. P. Singh (eds.), Panel surveys. New York: Wiley, pp. 319–339.