1
$\begingroup$

We conducted a study where we are measuring perceived tiredness at different intervals (a total of 10). We recorded a score before starting the experiment, as a covariate to account for differences in baseline (i.e. different people would have started at different levels of tiredness).

The experiment asked them to do a single task (keep pressing a button occurring at a random location on a smartphone). After every 1 minute, the activity (smartphone app) paused, and the participants were asked to give a score between 0-10 for their perceived tiredness. The activity ran for 10 minutes and thus we have 10 scores for each subject (plus 1 for the baseline).

I want to compare the effect of the condition (treatment) between all the participants (within-subject).

Now, ANCOVA is good for adjusting pre-treatment scores (i.e. baseline) and any variations that might come due to participants starting at different levels of tiredness.

However, ANCOVA is generally good for comparison between groups. I only have one group and everyone is doing the same tasks. I can treat every participant as a group (since I want to compare within participants), and run ANCOVA but I am not convinced this is the right approach.

Suggestions?

$\endgroup$
  • 1
    $\begingroup$ What do you mean that you "want to compare the effect of treatment", given that there is only one treatment & everyone got it? Do you want to compare some later measurement to the baseline? Which? Etc. $\endgroup$ – gung - Reinstate Monica Jun 16 '16 at 19:24
  • $\begingroup$ I meant, I want to measure the difference between the scores of each participant based on the condition (treatment) I gave them. $\endgroup$ – NewStatBoy Jun 16 '16 at 20:39
  • 1
    $\begingroup$ You must describe better how the study was organized? Were these tasks all done in sequence? Did you use a fully factorial design or latin squares? $\endgroup$ – AdamO Jun 16 '16 at 22:06
  • $\begingroup$ I added the details of the study. I hope that clarifies the question a bit more? Since it's only a single task with one DV, FFD and Latin Squares are not applicable. $\endgroup$ – NewStatBoy Jun 17 '16 at 0:11
  • $\begingroup$ All right, now I'm confused. I'd thought intervals were assigned between subjects, but now I see that each subject was measured 11 times on the tiredness measure. So what's the treatment? $\endgroup$ – Kodiologist Jun 17 '16 at 0:13
1
$\begingroup$

Try a linear mixed model with fixed effects for time, pose, and a time × pose interaction, and a batch of per-subject random effects. The fixed effects will then tell you the linear effects of condition and time, and the random effects will tell you each subject's bias in their tiredness ratings.

[My previous answer from when I misunderstood the design:] Try using a linear regression model with two predictors: interval length and baseline tiredness. The coefficient of interval length will then tell you the linear effect of interval when controlling for a linear effect of baseline tiredness. An even simpler approach is to subtract each subject's baseline tiredness from final tiredness and regress this on interval length.

$\endgroup$
  • $\begingroup$ How would the OP account for correlation of each subjects repeated measurements? I'd recommend a mixed effects model or using generalized estimating equations (GEE). $\endgroup$ – StatsStudent Jun 16 '16 at 21:52
  • $\begingroup$ @StatsStudent From my reading of the question, each subject experienced only one condition (that is, interval). Hence, there are only two measurements per subject, one at time 1 and one at time 2. $\endgroup$ – Kodiologist Jun 16 '16 at 22:11
  • $\begingroup$ -1 by someone? That's a bit harsh... (+1 because there is some usefulness in this approach but @StatsStudent raises a potential valid point - I think the post needs further clarifications anyway.) $\endgroup$ – usεr11852 says Reinstate Monic Jun 16 '16 at 23:41
  • $\begingroup$ Thank you for your response. I added details of the study to the question, which hopefully helps clarify it. @Kodiologist There are more than two measurements. $\endgroup$ – NewStatBoy Jun 17 '16 at 0:13
  • $\begingroup$ It's a minus one. Not a death sentence. $\endgroup$ – StatsStudent Jun 18 '16 at 22:53

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.