Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I am testing a hypothesis that states that participants after an intervention will be more likely to perceive themselves as being at risk for disease. Two questionnaires will be used. One will consist of 5 questions and will measure perceived susceptibility and the the other will consist of 7 questions and will measure perceived disease severity.

I think the best way though uncertain to analyze these two questionnaires which are rated on 5-point Likert scales is to use a $t$-test but I am not sure of how or the best way to do it. I am looking for suggestions.

share|improve this question
Now that I re-read the question and its title, it's not clear to me where the pre-post measurement process arises? Do you intend to use the two questionnaires twice (before and after intervention)? – chl Dec 26 '11 at 19:37
The answer is yes. I will be using the same instruments pre and post the intervention. – tiffiny Dec 27 '11 at 1:22
Are these questionnaires considered validated in your study population? Are there standard ways to score each (e.g., arithmetic sum of responses, weighted sum, different subscales)? – Ming-Chih Kao Dec 27 '11 at 5:48
I will have to do a pilot study initially. The scores will be tabulated and separated by the overall number of participants who answer each question to determine the score rates for perceived severity and susceptibility and additonal demographic questions. – tiffiny Dec 27 '11 at 15:35

T-tests assess a difference in mean outcome between two groups. In this case, I assume you intend to use your 5 level ordinal response scale as such an outcome. This means that responses will be literally coded as such with the 1 indicating a response of "not at all at risk" and 5 indicating "at very high risk". This is generally considered a valid approach for the analysis of such data.

Your study design allows you to use a special paired t-test for which software is available to compute and test for differences in pre/post responses. In a 1 sample case, without a control group, you might test whether the estimated pre/post difference is consistent with having no difference (a difference of 0). It's generally bad practice to do an intervention study without a control group. This gives rise to the Hawthorne effect.

If you have a control group, then you can use a 2 sample paired t-test to compare pre/post differences between the intervention and control groups. In a future study, it might be useful to consider adjusting for certain variables, like family history, for greater precision. This would require a regression framework.

share|improve this answer
Thank You I will definitely make use of your advice. – tiffiny Dec 27 '11 at 21:19
If I were to conduct this study using a regression framework what would you suggest ? I do intend to look at the varialbes BMI, family history of disease, age , personal history of disease. – tiffiny Dec 27 '11 at 21:24
It depends on lots of factors, most notably the specific disease. Take a peek at Regression Modelling Strategies or Regression Methods in Biostatistics which give rich discussions about variable adjustment in such models. – AdamO Dec 27 '11 at 22:15
Great thank you. – tiffiny Dec 29 '11 at 23:31

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.