Wednesday, May 26, 2010

Reply from author of stats paper.

Some time ago, I reported about a statistical analysis of stuttering risk factors: see here. They found no overwhelming risk factor, which is not surprising if genes and a specific brain insult plays a role. The lead author, Dr Vladeta Ajdacic-Gross, was so kind to answer my questions. Here they are:

1) How much of the variance can you explain with the found risk factors? (Incl. gender)
The Nagelkerke's R2 is 0.04, what was no surprise after excluding malingerers and SCL-high-scorers from the sample. Moreover, we found no strong risk factors in the analysis.
2) Did you look at other developmental disorders? I would be interested to know whether Tourette for example has similar risk factors.
We have used all potentially relevant variables which were available. Unforunately, information about Tourette's syndrome and several other issues of interest was lacking. We will have some space for a replication of this analysis under hopefully better conditions in a new large scale project we are preparing in psychiatric epidemiology.
3) How do you control for over-fitting issues due to the many variables? How many do you have?
Over-fitting is supposed to be a minor problem since the results are smoothed after sacrificing a great deal of variance (see point 1 above). Moreover, the frequencies in many cross-tabs (see
second column in Table 1) leave little space for overfitting issues. We have listed all variables we used in the analyses.
4) You only did this for the 2003 data, right? Why? Have you tried for up-to-now data? Would it change significantly the stat error?
The screening from 2003 was the first one to be run under real conditions and included an extended set of items. Screenings from 2004 and later years were based on reduced item sets, thus lacking also the stuttering item.
5) I am not sure about the foreign parent, because in Switzerland there are different languages in different Kantons. Do you mean different-language parents or Swiss non-nationals?
The information about different-language parents would have been preferable but was not available. In addition, a certain proportion of foreign parents are from German or Austrian origin. N detailed information was available.
6) The database might be used in conjuction with genetics work.
This is/was not possible at all. The Swiss Army had a very pragmatical approach to the screening. In addition, data security issues are generally of much greater relevance in Switzerland than potential scientific benefit.
7) Have you found differences in childhood stuttering and adult stuttering risk factors?
As mentioned in the article, we had no information about whether and when stuttering stopped.

No comments: