我正在尝试预测响应类型的可能性(4个级别:正确,不正确,过早,遗漏)。我有4个预测变量:试验(连续变量,因主题而异,但约为1:80),组(分类变量,3个级别:“ HI”,“ MID”,“ LI”)和ITI(分类变量) ,分为4个级别:“ 3000”,“ 5000”,“ 7000”,“ 9000”)。
我想我了解如何看待主要效果和相互作用,但是我在相互作用项中得到了一些奇怪的系数,这些系数并不能真正反映数据,所以我想知道自己是否做错了事。
我的数据
mydat1$rat_ID<- as.factor(mydat1$rat_ID)
mydat1$response_type <- factor(mydat1$response_type, levels = c("correct", "incorrect", "premature", "omission"))
mydat1$response_type2 <- relevel(mydat1$response_type, ref = "incorrect")
mydat1$ITI <- factor(mydat1$ITI, levels = c("3000", "5000", "7000", "9000"))
mydat1$impulsivity <- factor(mydat1$impulsivity, levels = c("HI", "MID", "LI"))
#this is what my data looks like
head(mydat1)
Corr Prem Omiss Incorr ITI rat_ID Trials response_type impulsivity response_type2
1 1 0 0 0 3000 1 1 correct MID correct
2 0 0 0 1 3000 1 2 incorrect MID incorrect
3 0 0 1 0 3000 1 3 omission MID omission
4 1 0 0 0 3000 1 4 correct MID correct
5 0 0 1 0 3000 1 5 omission MID omission
6 0 0 0 1 3000 1 6 incorrect MID incorrect
str(mydat1)
'data.frame': 6414 obs. of 10 variables:
$ Corr : num 1 0 0 1 0 0 0 0 0 0 ...
$ Prem : num 0 0 0 0 0 0 0 0 0 0 ...
$ Omiss : num 0 0 1 0 1 0 0 1 1 1 ...
$ Incorr : num 0 1 0 0 0 1 1 0 0 0 ...
$ ITI : Factor w/ 4 levels "3000","5000",..: 1 1 1 1 1 1 1 1 1 1 ...
$ rat_ID : Factor w/ 24 levels "1","2","3","4",..: 1 1 1 1 1 1 1 1 1 1 ...
$ Trials : int 1 2 3 4 5 6 7 8 9 10 ...
$ response_type : Factor w/ 4 levels "correct","incorrect",..: 1 2 4 1 4 2 2 4 4 4 ...
$ impulsivity : Factor w/ 3 levels "HI","MID","LI": 2 2 2 2 2 2 2 2 2 2 ...
$ response_type2: Factor w/ 4 levels "incorrect","correct",..: 2 1 4 2 4 1 1 4 4 4 ...
然后我运行多项式逻辑回归
test <- multinom(response_type2 ~ Trials*ITI*impulsivity, data = mydat1)
#coefficients of the logistic regression
summary(test)
z <- summary(test)$coefficients/summary(test)$standard.errors
##p values of the coefficients
p <- (1 - pnorm(abs(z), 0, 1)) * 2
p
例如LI的输出与HI进行比较,ITI9000与ITI3000进行比较
Coefficients:
Trials:ITI9000:impulsivityLI
correct 0.016879271
premature 0.046295381
omission 0.003091414
我的问题是互动的系数无法反映数据,例如事实并非如此,与不正确的响应相比,ITI9000vsITI3000中的LIvsHI试用增加了。
有人可以告诉我是否应该在此模型中添加一些内容吗?