Mice weighed on average 41·7 ± 8·9 g (*n* = 9) upon initial capture. There was a significant positive effect of body mass on DEE (*F*_{1,6} = 6·14, *P* = 0·048 and *F*_{1,6} = 19·46, *P* = 0·005 for control and reduced groups, respectively), but there was no effect of group size (*F*_{1,6} = 2·51, *P* = 0·164 and *F*_{1,6} = 0·09, *P* = 0·779 for control and reduced groups, respectively) on DEE. Therefore, it was not the case that mice in naturally smaller group sizes had significantly higher DEE values. DEE values increased by a mean of 19% (10·8 kJ day^{−1}) when group sizes were reduced by one-half (70·0 ± 18·1 and 59·1 ± 13·2 kJ day^{−1}, respectively, two-sample *t* = 2·57, *P* = 0·03, Fig. 2a). There was no effect of body mass or group size on the increase in DEE when groups were reduced (*F*_{1,6} = 3·58, *P* = 0·107 for body mass, *F*_{1,6} = 1·78, *P* = 0·230 for group size). Therefore, it was also not the case that the increase of DEE (upon reduction of group size) was correlated with original group size or the body mass of the measured individual. We were interested in whether we could detect any effects of natural variation in group size on DEE. Since including too many covariates in a model with small sample size diminishes statistical power (MacCallum, Browne & Sugawara 1996), we reanalysed the DEE data slightly differently by using each covariate (body mass and group size) separately, using GLM analyses. In the control group, we found significant positive and negative effects for mass (*F*_{1,7} = 11·31, *P* = 0·01) and group size (*F*_{1,7} = 5·84; *P* = 0·046), respectively. By comparison, although there was a significant effect of mass when groups had been reduced (*F*_{1,7} = 27·38, *P* = 0·001), there was no significant effect of group size (*F*_{1,7} = 3·84, *P* = 0·09). These results suggest that there might have been effects of natural group size on DEE but that our sample size was too small (*n* = 9 groups) to detect significant differences when both covariates were included in the same model. SusMS averaged 1·41 ± 0·19 in natural group sizes. There were no effects of body mass or group size on SusMS (*F*_{1,5} = 0·55, *P* = 0·49 for body mass and *F*_{1,5} = 2·42, *P* = 0·18 for group size), or any significant change in SusMS when group size was reduced.

Assuming that mice gain the benefits of huddling for the duration that they are in the nest (*c*. 14 h day^{−1} in the winter and 10·5 h day^{−1} in the summer: C. Schradin, personal observation), we calculated the average expected increase in DEE for mice in the free-living group sizes that we measured using the derived relationship between temperature and group size on *V*O_{2} (Equation 3). We calculated the difference in *V*O_{2} for the range of free-living group sizes that were measured for DEE and for those that had been reduced by half. We then converted the difference between these two values of *V*O_{2} to kJ day^{−1} using a factor of 20·51 kJ l^{−1} O_{2} (Hardy 1972). As this figure would indicate the energy increase due to a reduction in group size for 24 h, we converted this into an equivalent increase in energy expenditure for a 14-h period, which was the length of time that mice were in their nests (huddling) at night. Using this method, the calculated average increase in DEE for the free-living mice for which group size had been reduced by one-half was 9·3 ± 3·0 kJ day^{−1}.