logistf模型使用sjPlot::plot_model绘制带误差棒图形的方法咨询
解决方案
sjPlot的plot_model()未对logistf类对象做完整适配,无法自动提取置信区间生成误差棒,可通过以下三种方案解决:
方案1:用ggeffects包计算置信区间后绘图(最适配原需求)
ggeffects是sjPlot的底层依赖包,已适配logistf对象,可直接生成带置信区间的预测结果,绘图风格和plot_model()完全一致:
单变量预测概率图(匹配种族效果图需求)
library(ggeffects) library(ggplot2) library(sjPlot) # 计算race_grp的预测概率和95%置信区间 pred_df <- ggpredict(lr3, terms = "race_grp") # 绘图,自动带误差棒,和原plot_model效果完全对齐 ggplot(pred_df, aes(x = x, y = predicted, ymin = conf.low, ymax = conf.high)) + geom_point(size = 3) + geom_errorbar(width = 0.2) + theme_sjplot2() + scale_color_sjplot("simply") + labs(title= "预测的饥饿发生概率", x= "种族", y="百分比")
交互项效果图(匹配原type="int"的需求)
如果要绘制wave和ff_hcondhas的交互效应,修改terms参数即可:
pred_int <- ggpredict(lr3, terms = c("wave", "ff_hcondhas")) ggplot(pred_int, aes(x = x, y = predicted, color = group, ymin = conf.low, ymax = conf.high)) + geom_point(position = position_dodge(width = 0.3), size = 3) + geom_errorbar(position = position_dodge(width = 0.3), width = 0.2) + geom_line(position = position_dodge(width = 0.3)) + scale_color_sjplot("simply") + theme_sjplot2() + labs(title = "交互效应预测概率", x = "调查波次", y = "预测概率", color = "是否有健康状况")
方案2:直接给现有plot_model结果加误差棒
如果不想调整原有绘图逻辑,可直接在原plot_model输出的ggplot对象上叠加误差棒:
library(ggplot2) p <- sjPlot::plot_model(lr3, type="int", mdrt.values = "meansd", show.values = TRUE, value.offset = .3) # 给绘图数据补充置信区间,此处用Wald法计算,也可替换为logistf自带的profile似然置信区间提升准确性 p$data$conf.low <- plogis(p$data$predicted - 1.96*p$data$std.error) p$data$conf.high <- plogis(p$data$predicted + 1.96*p$data$std.error) # 叠加误差棒 p + geom_errorbar(aes(ymin = conf.low, ymax = conf.high), width = 0.2, position = position_dodge(0.3))
注:logistf自带的profile penalized似然置信区间比Wald法更准确,可通过confint(lr3)提取后转换为概率尺度使用
方案3:用effects包绘图
effects包也已适配logistf对象,一行代码即可生成带误差棒的效果图:
library(effects) # 绘制种族变量的效应图 plot(Effect(focal.predictors = "race_grp", mod = lr3), main = "预测的饥饿发生概率", xlab = "种族", ylab = "概率")
内容的提问来源于stack exchange,提问作者Farida El Kafrawy
相关产品推荐
相关产品推荐

