Statistical Impact of Truncated Sampling or Alteration of Administration: The Case of the Geriatric Depression Scale

I applaud the multiple regression analysis done by Feher et al. in examining the validity of the Geriatric Depression Scale.1 I endorse their conclusions. I understand why the authors decided to exclude extremely depressed (and extremely demented) subjects. The authors cautiously and correctly mentioned in their discussion and abstract that their conclusions must, therefore, be limited to mild/moderate depression and mild/moderate dementia. The purpose of this letter is to remind readers and future researchers that such truncations of sampling produce other limitations that must be weighed in the interpretation of the results. Specifically, the reporting of data from such samples gives misleading information on the norms of scales in the general population: lower means, standard deviations, and correlations. Feher et al's data provide an excellent example of this. They reported a GDS mean of 7.8 and a standard deviation of 5.38. Even in community samples, means are usually around 10 (the upper bound for the “normal” range), and standard deviations are larger due to the fact that non-truncated samples tend to be skewed to the right, with a few subjects scoring very high in the moderate to severely depressed range (21–30 points on the GDS). In samples of medical in-patients or elders in long term care, the means are even a little higher. Correlations (especially product-moment coefficients) between depression scales are similarly lowered by truncated sampling. Feher et al. reported only a +.58 between the GDS and the Hamilton. Although the former is a self-report form and the latter is observer rated, and the former concentrates on affective and cognitive features while the latter includes many somatic components, these two scales have become well established as valid measures for depression in later life. Previously reported correlations between these scales were .82,2 .72, .75,3 .78,4 .62, and .81.5 Furthermore, Feher et al. administered the GDS in “self-report format” (ie, paper and pencil). Most investigators who have studied paper and pencil versus oral examination formats for administering the GDS have recommended the latter.6-9 My own experience is that this is especially true for frail and cognitively impaired subjects. (To their credit, Feher et al. had examiners available to answer any questions, but this is not the same as oral administration). Anyone wishing a free copy of the GDS, complete with instructions for administration and scoring, may write to me.