Survey

Form of Data

Surveys produce a structured set of data that forms a variable-by-case grid. In the grid, rows typically represent cases and columns represent VARIABLES. The cells contain information about a case's ATTRIBUTE on the relevant variable.

QUESTIONNAIRES are widely used in surveys because they ask the questions in the same way of each person and thus provide a simple and efficient way of constructing a structured data set. However, the cells in the data grid can, in principle, be filled using any number or combination of data collection methods such as interviews, observation, and data from written records (e.g., a personnel file).

Although the rows in the data grid frequently represent people, they can represent different UNITS OF ANALYSIS such as years, countries, or organizations. Where these types of units of analysis are used, the information in each column (variables) reflects information appropriate to that type of analysis unit (e.g., characteristics of a particular year, country, or organization).

Methods of Analysis

A second characteristic of surveys is the way in which data are analyzed. One function of survey analysis is to describe the characteristics of a set of cases. A variable-by-case grid in which the same type of information is collected about each case simplifies the task of description. However, survey researchers are typically also interested in EXPLANATION—in identifying causes. This is achieved by examining VARIATION in the DEPENDENT VARIABLE (presumed effect) and selecting an INDEPENDENT VARIABLE (presumed cause) that might be responsible for this variation. Analysis involves testing to see if variation in the dependent variable (e.g., income) is systematically linked to variation in the independent variable (e.g., education level). Although any such COVARIATION does not demonstrate CAUSAL RELATIONSHIPS, such covariation is a prerequisite for causal relationships.

Unlike EXPERIMENTAL research, in which variation between cases is created by the experimenter, social surveys rely on existing variation among the cases. That is, survey analysis adopts a relatively passive approach to making causal inferences (Marsh, 1982, p. 6). A survey might address the question “Does premarital cohabitation increase or decrease the stability of subsequent marriage relationships?” It is not possible to adopt the experimental method in which people are RANDOMLY ASSIGNED to one of two conditions (premarital cohabitation and no premarital cohabitation) and then seeing which relationships last longer once the couples marry. The social survey adopts a passive approach and identifies couples who cohabited before marriage and those who did not and then compares their breakdown rates.

The problem in survey research is that any differences in marriage breakdown rates between the two groups could be due to differences between the two groups other than the differences in premarital cohabitation (e.g., age, religious background, ethnicity, race, marital history). Survey analysis attempts to control for other relevant differences between groups by statistically controlling or removing the influence of these variables using a range of multivariate analysis techniques (Rosenberg, 1968).

...