2
$\begingroup$

I have two groups and must conduct hypothesis testing by using an appropriate stats test. Seeing as I have 26 participants in each group, do not have the standard deviation (or any other information other than the data points for each group), I thought that the t-test would be the best choice.

null hypothesis = there is no significant difference between both groups alt hypothesis = such a difference exists

Is there another statistical test that could be performed (like the z test or Fmax)? Is the t-test the only hypothesis testing that I should be doing here, or could I add another one?

Thank you

$\endgroup$
7
  • $\begingroup$ What hypothesis do you wish to test? $\endgroup$
    – whuber
    Dec 1, 2014 at 0:59
  • $\begingroup$ The null hypothesis is that there is no significant difference between both groups, and the alt hypothesis is that there is such a difference. Is this specific enough? Thank you @whuber $\endgroup$ Dec 1, 2014 at 1:13
  • 1
    $\begingroup$ The word "significant" doesn't really belong in hypotheses, which are statements about populations. If you're interested in "no difference" vs "any kind of difference", that sounds more like a two-sample goodness of fit test. What are you actually trying to find out? What's the underlying question you're trying to address? $\endgroup$
    – Glen_b
    Dec 1, 2014 at 1:16
  • $\begingroup$ I'm afraid it is not sufficiently specific: you need to state in what way you want to measure any differences. Are you concerned about differences of means? Medians? Variances? Distributions? Something else? (Incidentally, a valid hypothesis can make no reference to "significance": that is a matter of how the hypothesis is evaluated.) $\endgroup$
    – whuber
    Dec 1, 2014 at 1:17
  • $\begingroup$ @Glen_b This is all for a research paper in which my study attempts to find out wether or not there's a significant difference between the failure rate of a sample of students who regularly watch Netflix and a sample of students who don't watch it at all. My data points represent the percentage of tests failed all throughout high school in both groups. I formed the hypothesis myself, and so the goal of this study. Could that be where the mistake is? How can I reformulate to make it work with a t-test? Thanks $\endgroup$ Dec 1, 2014 at 1:31

1 Answer 1

2
$\begingroup$

My data points represent the percentage of tests failed all throughout high school in both groups.

This looks like count data; you'll likely have variance changing as the mean changes, skewness, and possibly distinct discreteness effects if the percentage of tests failed is small.

A more suitable approach than t-tests would be one suited to count data - chi-squares or perhaps binomial GLMs.

This is all for a research paper in which my study attempts to find out wether or not there's a significant difference between the failure rate of a sample of students who regularly watch Netflix and a sample of students who don't watch it at all.
My data points represent the percentage of tests failed all throughout high school in both groups.

You might want to consider whether there are other important differences between the two groups.

$\endgroup$

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.