 |
 |
Hi Catherine, Please see below: On 12 September 2017 at 13:46, Catherine Mewborn <[log in to unmask]> wrote: > Dr. Winkler, > > Thank you for your reply. I have several follow-up questions which I think > will help me better understand your response. > > There were baseline differences in age and dementia status between the > experimental and control groups that I think might be biasing the results > of the intervention. So, in my analysis, I entered baseline values for age > and dementia status as "control variables" and I want to know if there are > effects of the intervention above and beyond any effects of age and > dementia status on FA values, for example, as these effects are fairly well > established. Again, since it was a complicated 2x2 ANOVA design, my outcome > variables are change scores for the DTI parameters (e.g., FA) which > represent any change from pre- to post-intervention, and were used in a > two-sample t-test. > > Based on this information, does the contrast I listed in my first post: 1 > -1 0 0 (with 1 -1 being the contrast of exp vs. control group and age and > dementia status being assigned values of 0) accurately model the variables > to give the desired information? In other words, would this contrast > reflect larger changes in the experiential group (1) than the control group > (-1), above and beyond any effect of age and dementia status on the change > scores? > > It's possible to use this design and contrasts, that is, a two-sample t-test on the differences, with age and dementia status as nuisance. It will address the confounding effect of having larger differences overtime because of age differences at baseline. However, this model implies that FA could quadratically according to age, and that may not be an accurate representation. I just wanted to let you know about this. > Again, I was expecting that controlling for baseline age and dementia > status would reduce the significant of the results, given that there were > significant baseline group differences in these two variables. However, the > results did not change very much, leading me to wonder whether the EVs > and/or contrasts were set up incorrectly. > > This could be because FA does not change quadratically according to age. If changes in FA are linear over time, then there is no need to enter it in the model, even if there were baseline differences in age, because the subtraction already takes care of this. I am also not sure I understand your statement that TFCE statistics would > always be positive. FA is positive, but if it decreased across the year in > the study, which I might expect with older adult populations, the change > scores would be negative, and I assumed the t-statistic would be negative > as well, reflecting the deterioration of FA. (I calculated all the change > scores as time 2 minus time 1). > Yes, the t-statistic can be negative but TFCE uses only the positive side of the map. If an effect is negative, to be able to see it with TFCE, you need to look into the opposite contrast (i.e., the one with the polarity reversed). Hope this helps! All the best, Anderson > > Thank you for your time and help! > > Best, > Catherine Mewborn > > > > >