Amelia包中EM Bootstrap方法调用及多重插补结果疑问

Hey there! Let's break down your questions step by step since you're new to multiple imputation—no worries, we've all been there.

1. How to call the EMB (EM Bootstrap) method in Amelia #

To use the EM Bootstrap imputation method in Amelia, you just need to specify the boot.type argument in the amelia() function and set it to "emb". This tells Amelia to use EMB instead of the default parametric bootstrap. You can also adjust the emb.size parameter if needed (it controls the number of bootstrap samples used in the EMB step, default is 1000). Here's an example:

Completed_data <- amelia(XNanData, m = 3, p2s = 0, boot.type = "emb", emb.size = 1000)

Note that EMB is particularly useful when your dataset is small, as it tends to perform better than standard bootstrap in those cases.

2. Is your imputation code correct, and is dataI a valid Amelia imputed dataset? #

Your code is totally correct! Let's break it down:

• amelia(XNanData, m=3, p2s=0) generates 3 imputed datasets (m=3) and disables predictive mean matching (p2s=0—that's fine if you don't need that feature).
• Completed_data$imputations is a list where each element is a full imputed dataset. So Completed_data$imputations[[3]] pulls out the third imputed dataset, which is absolutely a valid imputed dataset generated by Amelia II.

3. Is dataI the merged result of multiple imputation? #

No, dataI is just one single imputed dataset—you haven't done the merging step yet! Remember, multiple imputation has three core steps:

1. Impute: Generate multiple complete datasets (you've done this with amelia(), creating 3 datasets).
2. Analyze: Run your statistical analysis (like a linear regression, t-test, etc.) on each of the 3 imputed datasets separately.
3. Merge: Combine the results from each analysis into a single set of inferential statistics (coefficients, standard errors, p-values).

To merge the results in Amelia, you'll use the mi.meld() function. Here's a quick example using a linear regression:

# Step 1: Analyze each imputed dataset
model1 <- lm(y ~ x1 + x2, data = Completed_data$imputations[[1]])
model2 <- lm(y ~ x1 + x2, data = Completed_data$imputations[[2]])
model3 <- lm(y ~ x1 + x2, data = Completed_data$imputations[[3]])

# Extract coefficients and standard errors from each model
coefs <- rbind(coef(model1), coef(model2), coef(model3))
ses <- rbind(summary(model1)$coefficients[,2], summary(model2)$coefficients[,2], summary(model3)$coefficients[,2])

# Step 2: Merge the results with mi.meld()
merged_results <- mi.meld(q = coefs, se = ses)

# View the merged output
print(merged_results)

This merged_results object will give you the combined coefficients, standard errors, and the adjusted degrees of freedom for inference.

内容的提问来源于stack exchange，提问作者amjay