我想知道随着时间的推移(我的意思是年龄,而不是波浪)的体积发展是否在群体之间有所不同。我也有一些协变量。A做了一个模拟数据集:
library(simstudy)
def <- defData(varname = "volume", dist = "normal", formula = 1000,
variance = 100)
def <- defData(def, varname = "group", formula = "1;2;3", dist = "categorical")
def <- defData(def, varname = "age", dist = "normal", formula = 14, variance = 2)
def <- defData(def, varname = "scanner", formula = "1;2", dist = "categorical")
set.seed(1)
sim.data <- genData(220, def)
我想出了以下模型:
library(mgcv)
sim.data$group <- as.factor(sim.data$group)
m1 <- gamm(volume ~ group + s(age, by = group, k = 20, bs = "tp") + scanner , data = sim.data, random=list(id=~1))
summary(m1$gam)
这导致以下输出
Family: gaussian
Link function: identity
Formula:
volume ~ group + s(age, by = group, k = 20, bs = "tp") +
scanner
Parametric coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 997.3096 2.7303 365.280 <2e-16 ***
group2 0.5983 1.9557 0.306 0.760
group3 1.2819 1.7592 0.729 0.467
scanner 1.3806 1.3627 1.013 0.312
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Approximate significance of smooth terms:
edf Ref.df F p-value
s(age):group1 1 1 1.850 0.175
s(age):group2 1 1 0.001 0.969
s(age):group3 1 1 0.194 0.660
R-sq.(adj) = -0.0133
Scale est. = 2.1375e-07 n = 220
我希望我可以用我的 s(age,by=group) 术语来解释群体之间的发展差异,但是它为我提供了每个群体的平滑年龄参数。我怎样才能最好地评估交互效果(尽管我也在某处读到交互可能不是这个加法模型中的合适术语)?
我想到了使用 anova(),将带有 by term 的模型与没有 by term 的模型进行比较
m2 <- gamm(volume ~ group + s(age, k = 20, bs = "tp") + scanner , data = sim.data, random=list(id=~1))
anova(m1$gam,m2$gam)
但这并没有给我根据我将 lm 模型与 anova() 进行比较的经验所期望的输出
Family: gaussian
Link function: identity
Formula:
volume ~ group + s(age, by = group, k = 20, bs = "tp") +
scanner
Parametric Terms:
df F p-value
group 2 0.297 0.743
scanner 1 1.026 0.312
Approximate significance of smooth terms:
edf Ref.df F p-value
s(age):group1 1 1 1.850 0.175
s(age):group2 1 1 0.001 0.969
s(age):group3 1 1 0.194 0.660
有谁知道比较模型是否是研究交互效果以及如何获得实际效果的合适方法?最好是我也可以用于成对分析的方法。
所有的帮助将不胜感激!