Hi Greg, We are eager to know your opionion of the gap in the spread of the Language_Task_Story_Acc. Hope that is not a problem in the data.
Thanks, Jingwei On Thu, Sep 28, 2017 at 5:03 AM, Elam, Jennifer <[email protected]> wrote: > Hi Nicole, > > The emotion measures you mention are all from NIH Toolbox (TB), and they > did all the scoring calculations based on the entire TB national sample, so > you may want to ask them about the effect you see through their helpdesk at > [email protected]. > > > The Anger Physical Aggression, Loneliness, Instrumental Support, and > Perceived Rejection surveys measures you mention are very short > (5-8 item) fixed length surveys with 5-7 point scales for raw item answers > ranging from 1= "never" or "extremely untrue of me" to 5= "extremely true > of me" or "always". According to the NIH Toolbox Scoring and > Interpretation Manual that was current when we froze the HCP protocol > (attached), they are all scored with Item Response theory (IRT), which does > not assume items to be equally difficult and therefore does not > assume answers to be of equal value. The use of IRT might have > contributed to the irregular spread at one end of the scale. > > > Furthermore, the questions on these surveys measure more > extreme behavior/level of support (see pp. 226-277 of the attached TB Data > Dictionary for the questions asked) and we did have a lot of our normal, > healthy HCP young adult subjects answer these surveys in a way that put > them on one end of the scale (e.g. answered all items "never", or > "always"), which created an understandable floor or ceiling effect in the > data for these assessments. Looking at the raw item level data, subjects > with one item level answer 1 point away from the floor or ceiling jump to > the next value in the spread. > > > Greg, perhaps you can comment on the gap in the spread of the > Language_Task_Story_Acc? > My only comment is that the Story part of the Language task is relatively > easy, so a high number of subjects with 100% accuracy is expected. > > > Best, > > Jenn > > > Jennifer Elam, Ph.D. > Scientific Outreach, Human Connectome Project > Washington University School of Medicine > Department of Neuroscience, Box 8108 > 660 South Euclid Avenue > St. Louis, MO 63110 > 314-362-9387 > [email protected] > www.humanconnectome.org > > ------------------------------ > *From:* [email protected] <hcp-users-bounces@ > humanconnectome.org> on behalf of Nicole Kuek <[email protected]> > *Sent:* Tuesday, September 26, 2017 1:06:12 AM > *To:* [email protected] > *Subject:* [HCP-Users] Data outliers > > Dear HCP experts, > > In looking through some of the emotion and cognition behavioural data in > the hcp, we have noted some outliers that do not seem to be explained by a > floor or a ceiling effect. For example, AngAggr_Unadj, where a large group > of subjects have one much lower score, that is not continuous with the > spread of the other values (see attached figure AngAggr_Unadj). > > We have found this in a few of the emotion measures, both above and below > the spread (e.g Lonliness_Unadj, InstruSupp_Unadj, PercReject_Unadj, etc). > Please advise. > > Furthermore, should we be concerned about the gap between subjects scoring > 100 and 80+ in Language_Task_Story_Acc (see attached figure > Language_Task_Story_Acc)? > > -- > Best Regards, > > Nicole K. > > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
