Ngth. The correlation among FTR and the savings residuals was unfavorable
Ngth. The correlation in between FTR along with the savings residuals was damaging and important (for Pagel’s covariance matrix, r 0.9, df 95 total, 93 residual, t two.23, p 0.028, 95 CI [.7, 0.]). The results were not qualitatively various for the option phylogeny (r .00, t two.47, p 0.0, 95 CI [.eight, 0.2]). As reported above, adding the GWR coefficientPLOS One DOI:0.37journal.pone.03245 July 7,36 Future Tense and Savings: Controlling for Cultural Evolutiondid not qualitatively modify the outcome (r .84, t two.094, p 0.039). This agrees with the correlation identified in [3]. Out of 3 models tested, Pagel’s covariance matrix resulted within the very best match from the data, in line with log likelihood (Pagel’s model: Log likelihood 75.93; Brownian motion model: Log likelihood 209.8, FTR r 0.37, t 0.878, p 0.38; PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/25880723 OrnstenUhlenbeck model: Log likelihood 85.49, FTR r .33, t three.29, p 0.004). The fit on the Pagel model was significantly greater than the Brownian motion model 
(Log likelihood distinction 33.two, Lratio 66.49, p 0.000). The outcomes weren’t qualitatively distinct for the option phylogeny (Pagel’s model: Log likelihood 76.80; Brownian motion model: Log likelihood 23.92, FTR r 0.38, t 0.88, p 0.38; OrnstenUhlenbeck model: Log likelihood 85.50, r .327, t 3.29, p 0.00). The results for these tests run with the residuals from regression 9 aren’t qualitatively diverse (see the Supporting information and facts). PGLS within language families. The PGLS test was run within each and every language loved ones. Only 6 families had adequate observations and variation for the test. Table 9 shows the results. FTR did not considerably predict savings behaviour within any of those families. This contrasts with the results above, CAY10505 potentially for two reasons. Very first could be the challenge of combining all language families into a single tree. Assuming all households are equally independent and that all households possess the same timedepth is not realistic. This may perhaps mean that households that don’t fit the trend so properly may perhaps be balanced out by families that do. Within this case, the lack of significance inside families suggests that the correlation is spurious. Nevertheless, a second challenge is that the outcomes inside language families have a quite low number of observations and comparatively little variation, so might not have adequate statistical power. As an illustration, the result for the Uralic household is only based on three languages. In this case, the lack of significance within families may not be informative. The use of PGLS with a number of language households and with a residualised variable is, admittedly, experimental. We believe that the common idea is sound, but further simulation operate would have to be completed to work out whether or not it truly is a viable system. A single particularly thorny issue is how to integrate language families. We suggest that the mixed effects models are a improved test of the correlation between FTR and savings behaviour generally (plus the benefits of these tests suggest that the correlation is spurious). Fragility of data. Because the sample size is comparatively small, we would prefer to know no matter whether specific data points are affecting the result. For all data points, the strength of the relationship among FTR and savings behaviour was calculated whilst leaving that data point out (a `leave one particular out’ evaluation). The FTR variable remains substantial when removing any offered information point (maximum pvalue for the FTR coefficient 0.035). The influence of every point is often estimated making use of the dfbeta.