0
$\begingroup$

I have to do some statistical tests on a questionnaire and don't know which tests i need to do. The questionnaire have been marked as right or wrong and I have about 60 peoples results. The tests i need include;

  1. If different types of people have better knowledge (these are split into two groups, i.e. younger and older).

  2. Whether peoples knowledge increased after receiving information on the subject.

The questions are different on the before and after questionnaire and which one was given first has been randomly assigned.

I thought of doing unpaired t test for 1, a paired t test for 2, but for this type of data do i need to do a normality test? Should i even be doing parametric tests? I think these should be two tailed, but it would probably be assumed that if you give people information on a subject their knowledge would improve?

Thank you for any help.

$\endgroup$
  • $\begingroup$ You say that the questions have been marked as right or wrong, but are there multiple questions? So people have a score, of the number of questions they got right, from 0 to X. What's X? You say which questionnaire was given first was random - was it the same questionnaire? $\endgroup$ – Jeremy Miles Apr 10 '13 at 21:27
  • $\begingroup$ Hey, there are two parts to the questionnaire, 50% were given part a first and 50% part B first. Yes people have a score, for every correct answer they get a mark and this is out of 25. $\endgroup$ – user24174 Apr 10 '13 at 21:47
3
$\begingroup$

Some points:

  1. Each variable that defines a category or type ("young vs old", "male vs female") is a factor that influences the outcome (score on the questionnaire). You are doing a Factorial Experiment. You need to list all the relevant factors, their possible value, and mount an analysis using some software. First thing to look into is interaction between factors - for an example, usually ages makes math scores lower, except on Asians. The factor Race interacts with the factor Age, then. Then check homoscedasticity, then check normality. If all goes right, you can then proceed to ANOVA and / or t-tests.

  2. Giving information to the subjects makes it a paired-sample controlled trial or something like it :) The results of this t-test still are affected by the factors.

  3. Due to sample size (and some probable bias), I would go with presenting the information graphically and using non-parametric tests, like Kolmorogov-Smirnov to compare the two distributions.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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