I have searched, but I'm unable to understand: please help a beginner.

1) Scenario - 10 people (random sample) are asked 6 questions. The 6 questions are divided into 2 categories (i.e. 3 long words category questions and 3 short words category questions) which are the independent variable to test whether there is a difference between the long words category performance and the short words category questions (based on the word length effect on memory span for working memory). So the 10 subjects have individual scores based on long word performance and short word performance (all derived from being asked one questionnaire). I further go on and calculate the means for all the subjects: for example, the long word mean is 1.3 and the short word mean is 2.3. The standard deviations of the groups will also be calculated.

2) Question 1. To test if the two means from the same group with results based on different question categories of a single questionnaire administered in full to an individual, can I use t-test to establish if the means differ statistically (all material I have found talks about two different samples {normally as independent samples}, or a single sample but having undergone two separate tests {one before intervention then another after intervention, but the sample being referred to as dependent}? So I am not too sure about applying a t-test when the means are derived from one questionnaire and administered once to one group.

3) Question 2 - if I am to use a t test, should it be of independent or dependent?


1 Answer 1


Since your responses are on the same subjects, the answers to a given subject's short- and long- word answers are dependent. That is, it's a paired test.

On the information you give, it's not possible for us to judge whether all the assumptions of the t-test are reasonable (they won't be exactly true, but the t-test should still do okay if the samples aren't exactly normal for example). You'd need that the pair-differences are approximately normal (your sample size is small so you can't rely on large-sample effects at all); you don't necessarily need that the standard deviations for the short and long are close to equal.

I'd think that unless the distribution is really skew or the performance scores only take a few values, you're probably fine with a paired-t test. There are other alternatives.

  • $\begingroup$ thanks a lot, it really helped - really needed 2 kno if i could use a t-test and u answered the matter direct, i really appreciate it Nick (dont know if my previous comment displayed, but that just really helped with my assignment) $\endgroup$
    – kudzanayi
    Commented Aug 11, 2013 at 1:38
  • $\begingroup$ To be more specific, I clearly stated we don't have enough information to tell if you could use it, but that it would probably be okay. That doesn't mean it is okay. Please don't use text speak here, by the way, we're not texting but attempting to unambiguously impart information. Take pity on your helpers and take the trouble to spell out "to" and "know" and "you". $\endgroup$
    – Glen_b
    Commented Aug 11, 2013 at 4:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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