为什么effect（）和predict（）产生不同的模型预测？

此帖子的数据可用，R脚本和数据可用（R脚本也在下面的帖子中）。提前感谢您的帮助

我在

glmmTMB

中构建了一系列混合模型。我最好的两个模型如下

igm_20 <- glmmTMB(igm_pres ~  fRHDV2_arrive_cat + fseason + sage + save_ajust_abun + fseason*fRHDV2_arrive_cat + (1 | fsite), data = edit_pp_dat, family = binomial)

igm_21 <- glmmTMB(igm_pres ~  fRHDV2_arrive_cat + fseason + sage + save_ajust_abun + fseason*fRHDV2_arrive_cat + sage*fRHDV2_arrive_cat + (1 | fsite), data = edit_pp_dat, family = binomial)

（很抱歉提供了指向情节的链接，我没有足够的声誉发布实际情节）

从效应图中可以看出，两种模型中交互作用的影响非常相似，这并不奇怪。然后，我将这两个模型取平均值，如下所示：

mod_ave_list_1 <- list(igm_20, igm_21)
mod_ave_1 <- model.avg(mod_ave_list_1, rank = AICc)
summary(mod_ave_1)

predict_1 <- glmer(igm_pres ~  fRHDV2_arrive_cat + fseason + sage + save_ajust_abun + fseason*fRHDV2_arrive_cat + (1 | fsite), data = edit_pp_dat, family = binomial)

predict_2 <- glmer(igm_pres ~  fRHDV2_arrive_cat + fseason + sage + save_ajust_abun + fseason*fRHDV2_arrive_cat + sage*fRHDV2_arrive_cat + (1 | fsite), data = edit_pp_dat, family = binomial)

predict_list_1 <- list(predict_1, predict_2)
ave_predict <- model.avg(predict_list_1, rank = AICc)

a$sage <- c(0, 0, 0, 0, 0, 0, 0, 0)
a$save_ajust_abun <- c(0, 0, 0, 0, 0, 0, 0, 0)

然后，我创建了一个

newdata

框架，从中我制作并绘制了模型预测，作为生成与上述类似的

effect（）

绘图的一种方法。在进行模型预测时，我使用了数值预测的平均值，因为这是调用

effect（）

时发生的情况。我在

predict（）

函数中加入了

re.form=NA

，这样我得到了总体平均预测，因为我的模型包含随机效应

a <- as.data.frame(c("Summer", "Autumn", "Winter", "Spring", "Summer", "Autumn", "Winter", "Spring"))
a$fRHDV2_arrive_cat <- c("Pre-RHDV2 arrival", "Pre-RHDV2 arrival", "Pre-RHDV2 arrival", "Pre-RHDV2 arrival", "Post-RHDV2 arrival", "Post-RHDV2 arrival", "Post-RHDV2 arrival", "Post-RHDV2 arrival")
mean(edit_pp_dat$sage, na.rm = TRUE) #4.659477e-17
mean(edit_pp_dat$save_ajust_abun, na.rm = TRUE) #-3.004684e-17
a$sage <- c(4.659477e-17, 4.659477e-17, 4.659477e-17, 4.659477e-17, 4.659477e-17, 4.659477e-17, 4.659477e-17, 4.659477e-17)
a$save_ajust_abun <- c(-3.004684e-17, -3.004684e-17, -3.004684e-17, -3.004684e-17, -3.004684e-17, -3.004684e-17, -3.004684e-17, -3.004684e-17)
a$fsite <- c(NA, NA, NA, NA, NA, NA, NA, NA)
colnames(a) <- c("fseason", "fRHDV2_arrive_cat", "sage", "save_ajust_abun", "fsite")

predict.values <- predict(ave_predict, backtransform = TRUE, newdata = a, se.fit = TRUE, re.form = NA)

a$estimates <- predict.values$fit
a$se <- predict.values$se.fit
a$lci <- a$estimates - 1.96*a$se
a$uci <- a$estimates + 1.96*a$se
a$fseason <- factor(a$fseason, levels = c("Summer", "Autumn", "Winter", "Spring"))
a$fRHDV2_arrive_cat <- factor(a$fRHDV2_arrive_cat, levels = c("Pre-RHDV2 arrival", "Post-RHDV2 arrival"))

ggplot(a, aes(x = fseason, y = estimates, colour = fRHDV2_arrive_cat, group = fRHDV2_arrive_cat)) + geom_line(size = 1) + geom_point(size = 3) + geom_errorbar(aes(ymin = lci, ymax = uci), width = .2) + labs(x = "Season", y = "Predicted probability of IgM seropositivity", colour = "RHDV2 arrival category") + scale_color_manual(labels = c("Pre-arrival", "Post-arrival"), values = c("red", "blue")) + theme(axis.title.x = element_text(face = "bold", size = 16), axis.title.y = element_text(face = "bold", size = 16), axis.text.x = element_text(size = 14), axis.text.y = element_text(size = 14), legend.title = element_text(face = "bold", size = 14), legend.text = element_text(size = 12))

---

a对于那些可能感兴趣的人，我找到了解决问题的方法<代码>编辑数据$sage

和

编辑数据$save\u ajust\u abun

为标准化变量，因此其平均值为0。因此，

a$sage

和

a$save\u ajust\u abun

应如下所示：

mod_ave_list_1 <- list(igm_20, igm_21)
mod_ave_1 <- model.avg(mod_ave_list_1, rank = AICc)
summary(mod_ave_1)

predict_1 <- glmer(igm_pres ~  fRHDV2_arrive_cat + fseason + sage + save_ajust_abun + fseason*fRHDV2_arrive_cat + (1 | fsite), data = edit_pp_dat, family = binomial)

predict_2 <- glmer(igm_pres ~  fRHDV2_arrive_cat + fseason + sage + save_ajust_abun + fseason*fRHDV2_arrive_cat + sage*fRHDV2_arrive_cat + (1 | fsite), data = edit_pp_dat, family = binomial)

predict_list_1 <- list(predict_1, predict_2)
ave_predict <- model.avg(predict_list_1, rank = AICc)

a$sage <- c(0, 0, 0, 0, 0, 0, 0, 0)
a$save_ajust_abun <- c(0, 0, 0, 0, 0, 0, 0, 0)

a$sage