Test-retest reliability is a statement about how reliable the gathered information is by asking the same questions twice.
Talking about reliability is also talking about unreliability. If a line is drawn, set 0% on the left and 100% on the right. Now somewhere on this line, the researcher must determine where the reliability of his research is. One way to do this is to calculate test retest reliability.
As you can reed elsewhere in this dictionary, the data in the database are always slightly distorted, which is called bias. The bias can be caused by two reasons: the information is not submitted correctly or the information is not correctly recorded. This can be displayed in a formula:
With a test-retest the answers of both tests can be compared. Between the two scores a correlation coefficient can be computed. If the correlation coefficient is 1 then there is no distortion. This means, the information is very reliable. If the correlation coefficient decreases, the information gets unreliable. In the weird case that the coefficient is 0, the respondents give answers randomly.
Test-retest reliability cannot always be applied
It is not always possible to apply the test-retest reliability. Most questions cannot be asked twice.
In an interview the respondent will start questioning himself why you don’t believe him if you ask a second time for his age or opinion. He even might give different answers because he thinks you didn’t want to hear the answer he gave before. Now the opposite has been achieved of what the goal was. To apply test-retest in interviews the interviewer might summarize what has been said and then ask if he summarized it well.
In a questionnaire it is not done to ask the same question twice. At least, not at the same moment. It is possible to ask the same questions after a while. If the same answers are given, the questionnaire is reliable. However changes in the answers might be due to a change in opinion.
In case of construction an instrument, the respondent can be asked to cooperate in a research with this goal. To prevent the respondent knows what the research is about, he can be asked to fill in series of questionnaires of which some are the same. It might seem a waste of time and effort to delete the non-relevant information, but it provides a good insight of the quality of this test.