In our Commentary we made the (unoriginal) observation that the data in LOG13-blogs is remarkably skewed, making analysis difficult. In their reply (LGO15) Lewandowsky et al. are unmoved, stating that we ‘provided no justification to dismiss LOG13-blogs as its analysis recognized and accounted for the skew.’
No data processing method can ‘account for’ failing to survey the population that you are interested in. The LOG13-blogs survey was almost entirely completed by the ‘climate-convinced’, as shown below, which compares this group with the more representative LGO13-panel survey.
While the LOG13-blogs survey reached a small number of climate sceptics, the dearth of ‘climate moderates’ is particularly striking.
LGO15 says that an ‘ordinal rank-based’ analysis was used to correct for the skew. LOG13-blogs says that an ‘ordinal coding of the manifest variables, with the consensus responses binned into nine categories with approximately equal numbers’ was used in the analysis. However, since over 40 per cent of responses had the highest possible score for CLIM (and the response to any individual variable had only 4 possible values) the way in which the variables were actually binned is unclear.
This skew would not matter (much) if there was a straight-line relationship between CY and CLIM. It becomes problematic because of the non-linear relationship. This is clearly shown by the fact that if we remove the climate sceptics (11 % of the total, with CLIM < 2.0) from the LOG13-blogs dataset, we get a steeper straight-line correlation of CY with CLIM (although as already noted, the fit explains only a tiny percentage of the variance in both cases).
In other words, if no sceptics had been surveyed, their views would have appeared more extreme!