0
$\begingroup$

For my doctoral thesis work, I conducted a small feasibility study on the use of a virtual reality therapy protocol in preventing aggressive recidivism in 2 treatment-resistant schizophrenic patients. The study is prospective, observational, uncontrolled, and open. I administered most of the scales before and after the sessions, with two of them being administered weekly for the primary outcome measure. As a result, I have quantitative data resulting from the scores on these different scales. I would like to calculate the before-and-after difference in scores on these scales for the 2 patients, and I tried to perform a paired t-test without success because I have too little data. I've tested this on 3 different softwares (R, SPSS, JASP) without success. I'm a newbie and need help ! Do you have any suggestions on how to proceed? Thanks in advance !

$\endgroup$
3
  • 2
    $\begingroup$ What do you mean "too little data"? A t-test requires (at least) 2 points to work, how much data do you have? How much per group? What are you comparing the groups to? Do you have a control group? $\endgroup$ Sep 14, 2023 at 12:40
  • 2
    $\begingroup$ Can you be specific about what data you do have (eg. how many measurements for each of the two patients, how often, etc.) and what comparisons you want to perform? $\endgroup$
    – dipetkov
    Sep 14, 2023 at 12:51
  • $\begingroup$ Thanks for your responses. I have 14 measurements for each patient (14 measurements taken before the study and 14 after). I don't have a control group so I was hoping to just compare the before and after measurements. $\endgroup$ Sep 15, 2023 at 4:29

2 Answers 2

2
$\begingroup$

You should probably resist the temptation (or the pressure) to conduct a hypothesis test on feasibility study data, particularly with N=2.

A feasibility study has met its objectives if it tests the feasibility of a treatment or a trial procedure. You might also learn something about the outcome measures, or get a bound for the treatment effectiveness. But a hypothesis test should only really be used in a trial designed to test that hypothesis.

Also, while it's possible to conduct a paired t-test on two observations, I don't think you are likely to learn very much from it. If both of your participants happened to improve by the same amount over the treatment period you might get a small p-value, but would you be confident to attribute that to the treatment? And if there's a large p-value, it would tell you nothing beyond just describing the raw data.

I realise I am making some assumptions about what you are trying to do, so please correct me if I've misunderstood anything.

$\endgroup$
3
  • $\begingroup$ Thank you so much, it makes a lot of sense. Can I then only use a descriptive analysis to comment on the outcome measures ? $\endgroup$ Sep 15, 2023 at 4:39
  • $\begingroup$ Maybe. The study aims should have been in your ethics approval / protocol, so do whatever analysis is needed to meet those. From your comment you have 14 measures per patient per period, why did you take so many? Perhaps there is a research question that motivates taking all of these measurements? $\endgroup$ Sep 15, 2023 at 7:36
  • $\begingroup$ Indeed, the 14 data points were extracted from 6 questionnaires designed to assess various components of patients' social behavior (impulsivity, anger, aggressive behavior, social anxiety, emotional skills, etc.). We wanted to see if these various measures improved after the therapy protocol. $\endgroup$ Sep 15, 2023 at 8:57
0
$\begingroup$

I am guessing the score you are dealing with is the BPRS which uses questionnaire data and produces scores for 24 separate domains, scaled from 1 (no symptom) to 7 (severe). So whereas your dataset features N=48 pre-BPRS scores and another N=48 post-BPRS scores, they are not mutually independent. In your case, I believe you can summarize the data entirely by showing the number of domains showing an improvement from baseline, comparing the treated individual to control. Statistical significance testing would have, of course, no generalizability.

$\endgroup$
1
  • $\begingroup$ Thanks ! That is exactly what I was planning to do. The only problem being that I don't have a control group ... $\endgroup$ Sep 15, 2023 at 4:43

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.