Question

我正在使用nlme运行自启动的三参数逻辑模型，并希望阐明当模型中存在协变量时应如何选择起始值。

创建示例数据框：

sampledf <- tibble(
            age = rep(c(0, 3, 6, 9, 12, 15, 18, 21), 4), 
            mass = c(1, 4, 8, 12, 16, 20, 22, 23, 
            1, 5, 8, 13, 17, 21, 24.5, 24.5,
            1, 2.5, 6.5, 10, 14, 18.5, 20.5, 21,
            1, 5, 8, 13, 17, 21, 24.5, 24.5),
            treatment = c(rep(0:1, each=8, 2)),
            year = c(rep(0:1, each=16))) %>%
            slice(rep(1:n(), each = 3)) 

id = c("rae", "raphael", "reshmi", "rudy", "ruth", "roland", "remy", "roxy", "red", "randy", "raj", "roy")

sampledf <- sampledf %>% 
            arrange(age) %>%
            mutate(id = rep(id, 8),
            id = as.factor(id), year = as.factor(year), treatment=as.factor(treatment))

以图形方式显示S型曲线：

ggplot(sampledf, aes(age, mass, col=treatment, linetype=year)) + 
       geom_point() + 
       geom_smooth(method="gam", formula = y ~ s(x, k = 4))

因为该模型为每个参数取3个起始值{例如，Asym = c（23.0，26.1，24.5）}，这是否意味着列表中的前两个起始值应为治疗组的估计值，而第三，应该将两个年度的估计值一起平均吗？或者，第三个值应该是“治疗= 0，年= 0”的估计值吗？此外，在模型摘要中，假设每个参数有3个固定效果估算值（例如Intercept.Asym，Xmid.Asym，Scal.Asym），如何确定截距？即使没有估计值代表“治疗= 0，年= 1”，“治疗= 0，年= 0”组是否是参考点？

model1 <- nlme(mass ~ SSlogis(age, Asym, xmid, scal),
             fixed=list(Asym + xmid + scal ~ treatment + year),
             random = Asym + xmid ~ 1 | id,
             start=list(fixed=c(Asym=c(23.0, 26.1, 24.5),
                                xmid=c(9.33, 9.21, 9.1),
                                scal=c(3.63, 3.85, 3.7))),
             data=sampledf) 
summary(model1)

我如何确定要使用的起始值：

 sampledf %>% 
 group_by(treatment) %>% 
 do(fit = nls(mass ~ SSlogis(age, Asym, xmid, scal), 
            data = .)) %>% 
 tidy(fit) %>% 
 select(treatment, term, estimate) %>% 
 spread(term, estimate)

如何为nlme生长曲线选择自我起始值？

0 个答案: