Table 2. Summary of the mean maximum correlation (and SD among runs) achieved between measured/predicted height using windowed predictors (windowed = select the most significant SNP among those in a region).

Window size (kbp)	List size	Number SNPs included	GWAS	OLS
1000	4,026 (4)	4,026 (4)	0.375 (0.007)	0.465 (0.009)
500	7,975 (9)	7,975 (9)	0.358 (0.010)	0.496 (0.010)
200	19,516 (14)	19,516 (14)	0.344 (0.010)	0.535 (0.006)
150	25,758 (21)	20,000	0.343 (0.010)	0.545 (0.010)
100	37,839 (14)	20,000	0.340 (0.012)	0.556 (0.012)
50	71,869 (40)	20,000	0.341 (0.010)	0.580 (0.009)

Correlations should be compared with LASSO results in Table A1. The two methods used are called GWAS (use effect sizes from single SNP regression results), and OLS (using multi-SNP regression on windowed SNPs). Note that, while in agreement with the results presented in Wood et al. (2014), GWAS underperforms compared with OLS since it does not take correlations between SNPs into account beyond windowing. SNP: Single nucleotide polymorphism; OLS ordinary least-squares; LASSO: L1-penalized regression; GWAS: genome-wide association studies.