0
$\begingroup$

I have written and delivered a five-week programme aimed at reducing children's anxiety levels.

I tested participants' anxiety before and after the course using the Spence Anxiety Test. Spence standardises the raw data to typical children (t-scores, by age and gender).

I have also tested a 'random' group of similarly aged children from a general population. These kids have similar ages / gender mix, but have not been specifically selected for an anxiety course. Their mean anxiety t-scores are 50 (i.e. typical for their age/gender). The mean t-scores for the kids chosen for the course was 63 at the start of the course, indicating slightly elevated anxiety.

There were slight reductions in the average t-scores for both groups.

So far, I have data for 20 kids who have done the course and 20 'control' kids who did not do the course. I have their anxiety levels at week 1 and week 5 for both control group and subjects.

The questions:

1/ How do I test whether my programme has made a difference (statistically)? Is there a statistical difference between the change in anxiety in the two group?)

2/ How many sample points should I am for?

I'm not a statistician (as you can probably tell), but I do have some (rusty) knowledge of stats. I'm pretty nifty with Excel, which I will use for data analysis.

Any help greatly appreciated! Thank you.

Improve this question
$\endgroup$
2
  • $\begingroup$ Have you read through the following? Are you familiar with these designs? stats.stackexchange.com/questions/3466/… $\endgroup$
    – AdamO
    Commented Dec 19, 2017 at 18:44
  • $\begingroup$ I don't know if anything can reduce children's anxiety other than removing the source of the anxiety for each child specifically $\endgroup$
    – Aksakal
    Commented Dec 19, 2017 at 18:52

1 Answer 1

Reset to default
0
$\begingroup$

Looks to me like you need hypothesis testing! Your primary hypothesis would be that the course does not help, and you would show that your data supports nullifying it, under a model.

Hypothesis testing is addressed in more advanced stat classes. Here's one online resource:

http://www.randomservices.org/random/hypothesis/Introduction.html

A class in hypothesis testing should teach you:

  • how to choose a good model
  • how to measure effect size
  • how to estimate sample size, and the relation it has to effect size (the smaller the effect, the bigger the sample you need, in order to diminish variance)
  • how to choose a good p-value for your application
  • how to choose your hypothesis
  • and finally, how to test it, by combining your data, model, and p-value.

BTW, modern experiment design requires that you choose your model, hypothesis and method before running your experiment (which it sounds like you're already doing!). This is needed so that the researcher honestly runs his or her experiment and records the data without bias.

Imagine if a medical experimenter would choose more healthy looking patients to test a new medication, and more ill looking patients for the control group. That would obviously bias the result. He could do this even unconsciously. To avoid this bias, we use the above methodology to design a method to split the subjects randomly, before even seeing them.

Improve this answer
$\endgroup$

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.