Fidell

CHAPTER

Cleaning Up Your Act
Screening Data Prior to Analysis

This chapter deals with a set of issues that are resolved after data are collected but before the main data analysis is run. Careful consideration of these issues is time consuming and sometimes tedious; it is common, for instance, to spend many days in careful examination of data prior to running the main analysis that, itself,takes about 5 minutes. But consideration and resolution of these issues before the main analysis are fundamental to an honest analysis of the data. The first issues concern the accuracy with which data have been entered into the data file and consideration of factors that could produce distorted correlations. Next, missing data, the bane of (almost) every researcher, are assessed and dealt with.Next, many multivariate procedures are based on assumptions; the between your data set and the assumptions is assessed before the procedure fit is applied. Transforn~ationsof variables to bring them into compliance with requirements of analysis are considered. Outliers, cases that are extreme, create other headaches because solutions are unduly intluenced and sometimes distorted by them. Finally,perfect or near-perfect correlations among variables can threaten a multivariate analysis. This chapter deals with issues that are relevant to most analyses. However, the issues are not all applicable to all analyses all the time; for instance, multiway frequency analysis (Chapter 16) and logistic regression (Chapter lo), procedures that use log-linear techniques, have far fewer assumptions thanthe other techniques. Other analyses have additional assumptions that are not covered in this chapter. For these reasons, assumptions and limitations specific to each analysis are reviewed in the third section of the chapter describing the analysis. There are differences in data screening for grouped and ungrouped data. If you are performing multiple regression, canonical correlation, factoranalysis, or structural equation modeling, where subjects are not subdivided into groups, there is one procedure for screening data. If you are performing analysis of covariance, multivariate analysis of variance or covariance, profile analysis, discriminant analysis, or multilevel modeling where subjects are in groups, there is another procedure for screening data. Differences in these procedures areillustrated by example in Section 4.2. Other analyses (survival analysis and time-series analysis) sometimes have grouped data and often do not, so screening is adjusted accordingly. You may the material in this chapter difficult from time to time. Sometimes it is necessary find to refer to material covered in subsequent chapters to explain some issue. material that is more understandable afterthose chapters are studied. Therefore, you may to read this chapter now to get want an overview of the tasks to be accomplished prior to the main data analysis and then read ~t again after mastering the remaining chapters.

4.1

Important

Data Screeningin Issues
Data File of

4.1.1 Accuracy

The best way to ensure the accuracy of a data is to proofread the original file against thecomdata puterized data file in the data window. In SAS, data are most easily viewed in the Interactive Data Analysis window. With a small data file, is proofreading highly recommended, but with a large data file, it may not be possible. In this case, screening for accuracy involves of descriptive examination statistics and graphic of therepresentations variables. The first with a large step data toexamine is set univariate descriptive of through one statistics the descriptive programs such as SAS MEANSFREQUENCIES, or SPSS or UNIVARIATE or Interactive Data continuous For Analysis. are all the values within variables, range? Are means and standard deviations If you have discrete plausible? variables categories of religious affiliaas (such tion), are any there out-of-range numbers? you...