Griffiths KM, Christensen H.

Statistical review

This is a cross-sectional survey of 21 websites dealing with depression.

1) Abstract and elsewhere. It is unclear how the confidence intervals were calculated, and the level of confidence (eg 95%) is not given. The methods mention non-parametric tests but it is unclear whether these have been used for the confidence intervals. If the data are highly skewed, then conventional methods may be misleading (see Statistics with Confidence, Gardner MJ and Altman DG. BMJ publications)

2) P5. A significant correlation is not indication of good agreement, although a high correlation is. It would help to give the difference in scores between the two judges, with the s.d. to indicate level of variability in scoring.

3) On pages 10-11, there is too much emphasis on p-values. What is important is the difference, say, between the content score for sites with/without an editorial board.

4) In Table 3 what is the difference between ownership type, and ownership structure?

5) In Table 3, I am unsure of the value of quoting a p-value for one judge, when it had been agreed, a priori, to average the judges' scores for analysis.

6) In Table 3, the size of the s.d.s indicates that much of the data are highly skewed. I suggest that the s.d.s are omitted, but the mean score and range for the 21 sites included as a final row. For the important contrasts, the difference in scores and 95% CIs could be quoted in the text.

M J Campbell