Birte Reichstein
2010-Aug-05 14:41 UTC
[R] interpretation of summary.lm() for ANOVA and ANCOVA when dealing with 2 or more factors
Hi, I am having a hard time getting what the summary.lm-output for an ANOVA / ANCOVA means. Examples I find always seem to deal with simpler cases than what I meet in my data. My main problem is understanding the output when getting significant INTERACTION TERMS (what never occurs in examples :(). The following is the output after summary.lm(ancova) where "week" is continuous, "refug" is a 2-level (y,n) factor and "behand" is a 3-level (high,low,mix) factor. <Call: <lm(formula = length ~ week * refug * behand) < <Residuals: < Min 1Q Median 3Q Max <-3.91693 -0.53805 -0.01267 0.52403 4.04333 < <Coefficients: < Estimate Std. Error t value Pr(>|t|) <(Intercept) 6.94611 0.07175 96.811 < 2e-16 *** <week 0.77637 0.03894 19.938 < 2e-16 *** <refugy 0.15265 0.09633 1.585 0.1132 <behandlow 0.31196 0.12404 2.515 0.0120 * <behandmix 0.28687 0.12467 2.301 0.0215 * <week:refugy 0.30540 0.05312 5.749 1.04e-08 *** <week:behandlow 0.31952 0.06899 4.632 3.86e-06 *** <week:behandmix 0.16138 0.06717 2.403 0.0164 * <refugy:behandlow -0.27929 0.16623 -1.680 0.0931 . <refugy:behandmix -0.17837 0.17393 -1.026 0.3052 <week:refugy:behandlow -0.51495 0.09173 -5.614 2.26e-08 *** <week:refugy:behandmix -0.40225 0.09470 -4.248 2.26e-05 *** <--- <Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 <Residual standard error: 0.9312 on 1978 degrees of freedom < (2 observations deleted due to missingness) <Multiple R-squared: 0.5722, Adjusted R-squared: 0.5698 <F-statistic: 240.5 on 11 and 1978 DF, p-value: < 2.2e-16 In my understanding I get so far: Where there are only factor levels only the intercept is affected and where my continuous variable is involved the slope is affected. What I also understand is that the alphabetically first factor levels of my factors are somewhat hidden in the first two coefficients (intercept and week). BUT IN WHAT WAY ARE TWO DIFFERENT FACTOR_LEVELS HIDDEN IN THE SAME COEFFICIENT? AND WHAT ABOUT THE INTERACTION TERMS FOR THOSE TWO? Thanks for any help! Maybe there is a way to use contrasts in some way that would make the output easier to grab? /B [[alternative HTML version deleted]]