Page 368 - Contributed Paper Session (CPS) - Volume 6
P. 368
CPS1969 Janna M. De Veyra
Logit[(X)] = + + +
1 1
2 2
where ⍺, , and are estimates obtained in the procedures used and is the
2
1
random error. In this paper, the random error is assumed as normally
distributed with mean 0 and variance 3 in imputing z value in source A and
normally distributed with mean 0 and variance 4 in imputing y value in source
B. Same as with the regression imputation, missing values will be imputed
based on the probability of success.
2.2 Bootstrap Procedure
Bootstrap in this paper could be done in two ways. It could be done by either
resampling within the synthetic data or resampling across the synthetic data.
Resampling within synthetic data refers to resampling with replacement from
a combined data source where estimates on the missing data are already
available while resampling across the synthetic data refers to resampling with
replacement from a combined data source where estimates on the missing
data are not yet available. Both procedures were considered in this paper to
check for a possible difference in the result. Resampling would be done 200
times in both procedures where the number of samples in each pseudo
sample is just the same as with the original sample
2.3 Evaluation of the Procedure
The performance of the proposed procedures in testing for the independence
of statistically matched categorical variables was evaluated according to the
following:
1. To evaluate if the proposed procedures correctly reject the null
hypothesis, the power of the test will be computed by simulating data set
200 times wherein each replicate has a Chi-square statistic that is greater
than the critical value at a 0.05 level of significance. The proposed
procedures will then be applied and the Chi-square statistics in the
applied procedures will be computed in each replicate. The computed
power will be the total number of replicates with Chi-square statistics
obtained from the proposed procedures that is greater than the critical
value at a 0.05 level of significance divided by the total number of
replicates.
2. To evaluate if the proposed procedures falsely reject the null hypothesis,
the size of the test will be computed by simulating data set 200 times
wherein each replicate has a Chi-square statistic that is less than the
critical value at a 0.05 level of significance. The proposed procedures will
then be applied and the Chi-square statistics in the applied procedures
will be computed in each replicate. The computed size will be the total
number of replicates with Chi-square statistics obtained from the
357 | I S I W S C 2 0 1 9