My question is as follows:

I am using transcranial magnetic stimulation (TMS) to assess whether applying TMS for 10 minutes, 20 minutes, sham (not applied), active control (applied on brain site not responsible for the task) will impact object naming, and action naming. Furthermore, The TMS is applied every week for three weeks and the object naming and action naming is assessed each week.

The DV are object naming accuracy, action naming accuracy, object naming reaction time, action naming reaction time.

Its a randomized control trial so different participants are in different conditions of TMS. The questions I am asking: 1. Does TMS for 10, 20 minutes, sham, and active control impact my DV? Which condition has the largest effect? 2. Does the impact on the DV change after 1 week, 2 weeks, three weeks

Hope this makes sense.

  • 2
    $\begingroup$ Welcome to the site. Acronyms should not be used without defining them, especially when they are unrecognizable and not related to the field in which they are cited. You question needs to be better defined. What is your hypothesis, how are you measuring outcomes? $\endgroup$
    – Carl
    Commented Oct 5, 2016 at 14:56
  • $\begingroup$ Thanks you and apologies for using acronyms. TMS stands for trans cranial magnetic stimulation. My outcomes are picture naming, a verbal memory task, a working memory task. They are all behavioral measures. My hypothesis is that applying TMS for 20 minutes will have the most impact on all the tasks compared to the other conditions. Furthermore, there should be no effect of the sham and the active control TMS on the outcome measures. Hope this clarifies. $\endgroup$
    – Novice
    Commented Oct 5, 2016 at 16:01
  • $\begingroup$ OK, for our purposes we need to know how the outcomes are scored, e.g. on a 0 to 10 scale, integers only, or what? That determines how the outcomes can be analyzed. Each measurement's scale needs to be defined to understand exactly what is being asked. Regarding power, that depends largely on how well the measurements classify the results. For example, with a perfect classifier, one patient result is enough, that result would be classified 100% correctly. With a useless classifier, there is no number of trials that will be enough. $\endgroup$
    – Carl
    Commented Oct 5, 2016 at 16:20
  • $\begingroup$ Thank you! This question is not for a HW or test but an experiment I am planing to conduct. I am planning to measure the outcome as a continuous variable. $\endgroup$
    – Novice
    Commented Oct 5, 2016 at 17:42
  • 1
    $\begingroup$ I am scoring them two ways: accuracy and reaction time. For accuracy how much of the task do they get correct (e.g., 10 out of 20-) and for reaction time how fast are they in completing the task in milliseconds. $\endgroup$
    – Novice
    Commented Oct 5, 2016 at 18:26

1 Answer 1


I'm not sure what TMS has to do with all of this but anyway. Before thinking about an appropriate statistical test I would want to clarify my hypotheses. It sounds like you are just interested in whether your treatment is doing anything to your outcome measures ("impact my DV") which is probably not ideal (although I realize this is what many researchers do) unless your research is truly exploratory.

However, given that there are numerous findings suggesting that your treatment, depending on the stimulation parameters, has excitatory or inhibitory effects I'm assuming that you could (and probably should) develop a directional hypothesis first that predicts a specific effect of a specific direction. Ideally, this should also include an estimate of the effect size that you'd expect which could be informed (in best case) by meta-analyses. I think these steps are crucial in order to avoid the confusion between exploratory and confirmatory research which may lead to the unfortunate inflation of false positives by exploiting what has been termed researchers degrees of freedom. In addition, I would want to clarify my primary outcome measures. It seems that you are interested in naming accuracy (and reaction time); it sounds like you are going to test this in a task where objects and actions have to be named so I think this is really just one outcome measure that is tested under different conditions (objects and actions).

Once this is all clarified I would probably analyze the dataset with a multilevel model (linear mixed model) which gives you maximal flexibility in modeling the dependency between repeated measures and allows you to take all data into account without the need to eliminate entire cases based on missing observations. Which is another issue that should be addressed: do you expect missing observations? Assuming that your research might take place in a clinical (neuropsychological?) context (rTMS over several weeks to assess benefits on naming accuracy) I would expect that drop outs are likely to happen. Anyway, a multilevel model with a factor for treatment (four levels; 10 min, 20 min, sham, control-site), a factor for condition (two levels; object, action) and a factor for time (four levels; baseline, week1, week2, week3) as well as their respective interactions might be a start. However, you will obviously need to make sure that your study has enough power to detect the 3-way interaction implied in this model. In addition, I would want to include a random intercept per subject as well as random slopes for condition and time. This might not be easy to estimate with a small sample size but it is important to keep the random part of your model maximal (as the saying goes).

  • $\begingroup$ thank you so much for the answer and the tip regarding have a clearer hypothesis. $\endgroup$
    – Novice
    Commented Oct 18, 2016 at 16:44

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.