Hi,
I want to combine multiple amelia runs, and use the zelig function on the entire set of amelia outputs. However, I noticed that when I run zelig on the combined set, the coefficients don't look right. Here are the script and the R outputs:
> require(Amelia)
## Amelia II: Multiple Imputation
## (Version 1.7, built: 2013-02-10)
> library(Zelig)
Attaching package: ‘zoo’
The following object(s) are masked from ‘package:base’:
as.Date, as.Date.numeric
ZELIG (Versions 4.1-3, built: 2013-01-30)
> data(freetrade)
> a.out1 <-amelia(freetrade, m = 1, ts = "year", cs = "country")
-- Imputation 1 --
1 2 3 4 5 6 7 8 9 10 11 12 13 14
> a.out2 <-amelia(freetrade, m = 1, ts = "year", cs = "country")
-- Imputation 1 --
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
> a.out3 <-ameliabind(a.out1, a.out2) # combine the amelia runs
> summary(a.out3)
Amelia output with 2 imputed datasets.
Return code: 1
Message: Normal EM convergence
Chain Lengths:
--------------
Imputation 1: 14
Imputation 2: 15
Rows after Listwise Deletion: 96
Rows after Imputation: 171
Patterns of missingness in the data: 8
Fraction Missing for original variables:
-----------------------------------------
Fraction Missing
year 0.00000000
country 0.00000000
tariff 0.33918129
polity 0.01169591
pop 0.00000000
gdp.pc 0.00000000
intresmi 0.07602339
signed 0.01754386
fiveop 0.10526316
usheg 0.00000000
# run zelig on the first amelia run
> z.out1 <- zelig(tariff ~ polity + pop + gdp.pc, data = a.out1$imputations, model = "ls", cite = FALSE)
> summary(z.out1)
Call:
lm(formula = formula, weights = weights, model = F, data = data)
Residuals:
Min 1Q Median 3Q Max
-29.686 -11.067 -3.367 10.003 45.556
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.317e+01 1.891e+00 17.538 < 2e-16 ***
polity -1.060e-01 2.434e-01 -0.435 0.664
pop 2.925e-08 5.405e-09 5.411 2.15e-07 ***
gdp.pc -2.738e-03 5.235e-04 -5.231 4.99e-07 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 16.62 on 167 degrees of freedom
Multiple R-squared: 0.3234, Adjusted R-squared: 0.3112
F-statistic: 26.6 on 3 and 167 DF, p-value: 4.07e-14
# run zelig on the second amelia run
> z.out2 <- zelig(tariff ~ polity + pop + gdp.pc, data = a.out2$imputations, model = "ls", cite = FALSE)
> summary(z.out2)
Call:
lm(formula = formula, weights = weights, model = F, data = data)
Residuals:
Min 1Q Median 3Q Max
-40.13 -11.72 -3.63 10.16 43.77
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.400e+01 1.984e+00 17.133 < 2e-16 ***
polity -6.123e-02 2.545e-01 -0.241 0.81
pop 3.289e-08 5.689e-09 5.781 3.56e-08 ***
gdp.pc -3.016e-03 5.501e-04 -5.484 1.51e-07 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 17.47 on 167 degrees of freedom
Multiple R-squared: 0.3495, Adjusted R-squared: 0.3378
F-statistic: 29.91 on 3 and 167 DF, p-value: 1.574e-15
# run zelig on the combined amelia runs
> z.out3 <- zelig(tariff ~ polity + pop + gdp.pc, data = a.out3$imputations, model = "ls", cite = FALSE)
> summary(z.out3)
Call:
lm(formula = formula, weights = weights, model = F, data = data)
Residuals:
Min 1Q Median 3Q Max
-29.686 -11.067 -3.367 10.003 45.556
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.317e+01 1.891e+00 17.538 < 2e-16 ***
polity -1.060e-01 2.434e-01 -0.435 0.664
pop 2.925e-08 5.405e-09 5.411 2.15e-07 ***
gdp.pc -2.738e-03 5.235e-04 -5.231 4.99e-07 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 16.62 on 167 degrees of freedom
Multiple R-squared: 0.3234, Adjusted R-squared: 0.3112
F-statistic: 26.6 on 3 and 167 DF, p-value: 4.07e-14
Can someone please tell me why summary(z.out3) is returning the exact same results as summary(z.out1)?
How can I fix this problem so that the ameliabind output is treated the same as a direct output from the amelia function?
Thanks.
Hi all,
Craig Enders, in his Applied Missing Data Analysis (Guilford Press, 2010),
suggests that, if your regression model includes interaction terms, such
terms must also be included in the imputation model as a way to preserve
interaction effects (p. 265). Is there a way to set an interaction term
*within* Amelia code or should I generate interactions before the
imputation?
Suppose I am interested in the differential effect of education on
attitudes for different racial groups (say, African-Americans, Asians,
Latinos, and Whites). Can I set the ethnicity*education interaction
straight into Amelia code or should I create different variables for
African-Americans*education, Latino*education, and so on to catch the
race*education interaction?
Another possible question: Shouldn't I (or is it not necessary to) include
interactions in the imputation?
Best,
Fabricio.
