Step 9. Screen the data before carrying out analysis.

Data screening eliminates errors in the data. These errors may be caused by faulty entry by staff of data into a database . For example, the wrong keys may have been punched, or the data may have been entered into the wrong row or column. In the case of questionnaires and assessments, many of these keypunching errors can be eliminated by online administration, followed by automatic entry of the responses into a database for statistical analysis.

Errors may also be caused by respondents making unintended mistakes. Here are three common examples:

• The respondent may fail to skip certain questions s/he was supposed to skip.
• The respondent may give an "out-of-range" response: for example, when asked how many hours they work per week, they write "95," or, when asked their age, they write "125."
• The respondent may give inconsistent responses relative to the logic of the questionnaire. An example would be a teacher who reports that she has taught for only three years but then reports that she has taught with a particular classroom resource for five years.

Errors like these can be detected in various ways. For example, statistical software packages can generate counts (otherwise known as "frequencies") of the different values for each variable in the data set. They can also generate specific values of variables identified for each respondent, which can be used to detect an erroneous response. In some spreadsheet programs and statistics programs, you can also program "if/then" commands that will allow you to quickly identify erroneous values, such as responses to questions that a particular respondent should have skipped.