在附加分类变量 IndVar4(因子 a、b、c)的情况下,系数表可以不仅仅是一行更长的时间。添加变量 IndVar4:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.7548180 1.4005800 -1.2529223 0.2102340
IndVar1 -0.2830926 1.2076534 -0.2344154 0.8146625
IndVar2 0.1894432 0.1401217 1.3519903 0.1763784
IndVar3 0.1568672 0.2528131 0.6204867 0.5349374
IndVar4b 0.4604571 1.0774018 0.4273773 0.6691045
IndVar4c 0.9084545 1.0943227 0.8301523 0.4064527
最大行数小于所有变量 + 所有类别:
max_values <- length(ind_vars) +
sum(sapply( dfPRAC, function(x) pmax(length(levels(x))-1,0)))
所以新的修正函数是:
p_values <-
cbind(formula_vec, as.data.frame ( do.call(rbind,
lapply(glm_res, function(x) {
coefs <- coef(x)
rbind(c(coefs[,4] , rep(NA, max_values - length(coefs[,4]) + 1)))
})
)))
但结果不如连续变量那么干净。我认为 Metrics 将每个分类变量转换为(levels-1)虚拟变量的想法给出了相同的结果,并且可能更清晰的呈现。
数据:
dfPRAC <- structure(list(DepVar1 = c(0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1,
1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 0, 1), DepVar2 = c(0, 1, 0, 0,
1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1),
IndVar1 = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1,
0, 0, 0, 1, 0, 0, 0, 1, 0),
IndVar2 = c(1, 3, 9, 1, 5, 1,
1, 8, 4, 6, 3, 15, 4, 1, 1, 3, 2, 1, 10, 1, 9, 9, 11, 5),
IndVar3 = c(0.500100322564443, 1.64241601558441, 0.622735778490702,
2.42429812749226, 5.10055213237027, 1.38479786027561, 7.24663629203007,
0.5102348706939, 2.91566510995229, 3.73356170379198, 5.42003495939846,
1.29312896116503, 3.33753833987496, 0.91783513806083, 4.7735736131668,
1.17609362602233, 5.58010703426296, 5.6668754863739, 1.4377813063642,
5.07724130837643, 2.4791994535923, 2.55100067348583, 2.41043629522981,
2.14411703944206)), .Names = c("DepVar1", "DepVar2", "IndVar1",
"IndVar2", "IndVar3"), row.names = c(NA, 24L), class = "data.frame")
dfPRAC$IndVar4 <- factor(rep(c("a", "b", "c"),8))
dfPRAC$IndVar5 <- factor(rep(c("d", "e", "f", "g"),6))
设置模型:
dep_vars <- c("DepVar1", "DepVar2")
ind_vars <- c("IndVar1", "IndVar2", "IndVar3", "IndVar4", "IndVar5")
# create all combinations of ind_vars
ind_vars_comb <-
unlist( sapply( seq_len(length(ind_vars)),
function(i) {
apply( combn(ind_vars,i), 2, function(x) paste(x, collapse = "+"))
}))
# pair with dep_vars:
var_comb <- expand.grid(dep_vars, ind_vars_comb )
# formulas for all combinations
formula_vec <- sprintf("%s ~ %s", var_comb$Var1, var_comb$Var2)
# create models
glm_res <- lapply( formula_vec, function(f) {
fit1 <- glm( f, data = dfPRAC, family = binomial("logit"))
fit1$coefficients <- coef( summary(fit1))
return(fit1)
})
names(glm_res) <- formula_vec