# Science“驴得水”论文中的统计错误

2021年4月21号，Science发表了一篇论文，该研究发现，美国亚利桑那沙漠中的 “野驴挖井”现象会使起来动物渔翁得利，也能利用这些“井”中的水资源，进而有缓解生物多样性的丧失(Lundgren et al. 2021)

We then used generalized linear mixed effect models in the R package ‘glmmTMB’ v1.0.1 to analyze how resource types (i.e. equid wells, background waters, and dry controls) and environmental variables influenced vertebrate activity. We chose distributions appropriate for each response variable and nested date within site as a random effect. …we used zero-inflated mixed effect models with season nested within site as a random effect

richness.cels.glmmad.poisson <- glmmTMB(richness ~  water +  tmax + last_3_ppt +  (water:tmax)+ (water:last_3_ppt)+(1|numeric_day/site), data=all_water.div, ziformula = ~1, family=poisson)

OK，那么我们可以对其模型随机效应结构进行修正如下：

water +

tmax +

last_3_ppt +

(water:tmax)+

(water:last_3_ppt)+

(1|site/numeric_day),

data=all_water.div,

ziformula = ~1,

family=poisson)

可见，自由度不变，修正后模型的AIC值足足比原模型低了50个单位！模型改进之大，只能用铁血丹心来形容。同时，我们也可以比较两模型的R2

par(mar = c(5, 5, 2, 2))

plot(rd1$Estimate, rd2$Estimate,xlab="原模型（嵌套错误）参数估计值", font=2,

ylab="嵌套改正后模型参数估计值",cex.lab=1.5,

pch = 16, cex=1.5, ylim = c(-1,0.5), xlim=c(-1,0.5))

abline(a =0, b = 1, col = "blue")

v1<-1.168e-10

v2<-1.700e-10

v3<-2.672e-02

v4<-2.229e-10

points(x=c(v1,v2), y=c(v3,v4), pch = 16, cex=1.5, col="red")

Lundgren, E. J., D. Ramp, J. C. Stromberg, J. Wu, N. C. Nieto, M. Sluk, K. T. Moeller, and A. D. Wallach. 2021. Equids engineer desert water availability. Science 372:491-495.

https://blog.sciencenet.cn/blog-3442043-1362371.html

## 全部精选博文导读

GMT+8, 2024-2-26 16:52