go home

Select a Professional Development Module:
Key Topics Strategy Scenario Case Study References

Introduction |  Step 1 |  Step 2 |  Step 3 |  Step 4 |  Step 5 |  Step 6 |  Step 7 |  Step 8 |  Step 9

Step 9. Screen the data before carrying out analysis.

You will use clerical staff on your team to keypunch the questionnaire data onto computer tables using a common spreadsheet program. Then, you import the data into a computer-based statistics program. Each respondent questionnaire has its own ID, which makes it possible to locate the error in the data set. Once you have the data in the statistics program, you program a set of if/then rules into the statistics program to locate the errors.

You will look for two types of errors: keypunching errors and respondent errors. You discover keypunching errors. The questionnaire was constructed so that no responses should have been keyed in as 8s or 9s, yet your frequencies (otherwise known as "counts" in some statistics programs) reveal several of these values. You locate the 8s and 9s in the data set, identify the respondent questionnaires that the 8s and 9s would have come from and learn that the respondents made appropriate selections but the clerical staff keypunched them into the spreadsheet table improperly. You ask your clerical staff to make the corrections on the spreadsheet.

Further analysis reveals errors where respondents selected contradictory answer choices about the use of technology in their classrooms. For example, the values from two different items keypunched into the data set indicate a contradiction between the response to an item that reports little use of of technology with students, and a response to another item reporting frequent assignments of students to do in-class Internet research projects. To determine whether these and other contradictory responses are caused by respondent or the clerical staff, you obtain the respondent's questionnaire and learn that the errors were made by the respondents. You then examine all the responses from that respondent and find more errors. Hence, you conclude that you cannot trust any of these contradictory responses and remove them from the data set.