Analyzing Survey Data
Survey data is usually
analyzed in several stages using a variety of statistical procedures.
Generally, they fall into one of three broad categories.
There are a wide range of statistics that are used to describe or
summarize data. Most tables in survey reports present descriptive
statistics for one or more groups or sub-groups among the respondents.
For surveys, these usually are limited to either the the various percentages
of respondents who select from a range of categorical alternatives (e.g.,
"Very Dissatisfied" to "Very Satisfied", women v. men, position in the
company, etc.) or as averages (mean) or median (and percentile) scores on
items. Descriptive statistics computed for sample data (survey
respondents) are used as estimates for the target population (e.g., all
customers or employees) and as such, they should be accompanied by estimates
of the margin of error or of the confidence interval that give some
indication that is useful for gauging how accurate these descriptions are.
Usually, tests for significant differences are conducted to
examine changes that occur over time or differences between groups or
subgroups. Which specific procedures are used will depend on the type of
measurement scale used for the survey items under scrutiny. Differences
in categorical data, for example, may be analyzed using a procedure called Chi2
and differences in average (mean) scores may be analyzed using procedures such
as the t-test or the analysis of variance. By consensus, differences
that are unlikely to be the result of chance fluctuations in the data alone
(defined as a less than 0.05 probability) are referred to as statistically
significant. Alternatively, you can also examine margins of error or
confidence intervals for two groups or subgroups; if they do not overlap, the
differences are likely to be statistically significant using other tests.
There are a variety of procedures from simple correlations between items to
more elaborate statistical techniques that allow researchers to evaluate how
items relate to other items, individually or jointly. Regression
analysis and more complex statistical modeling procedures are based on the
intercorrelations among items and can be used to identify clusters of items
that fall together into single dimensions (e.g., which items appear to reflect
loyalty?) and to relate these items or dimensions to desired outcomes (e.g.,
how do satisfaction with compensation, opportunities for advancement, and the
effectiveness of organizational communications and other factors relate to
employee loyalty and commitment?). If you have a large enough sample
(generally a few hundred or more respondents) these techniques can help you to
develop a useful model of how your business objectives can be achieved.