Page 28 - Contributed Paper Session (CPS) - Volume 7

CPS2020 Honeylet T. S.
RBIAS decreases. Also, the RBIAS under these two matching methods are lower
compared to those of random hot deck imputation.
The three estimation procedures yield similar results under Poisson
regression and MCMC imputation matching methods, although Poisson
regression without bootstrap and bootstrap within methods yield the lowest
RBIAS. Poisson regression without bootstrap produces the lowest RBIAS when
the true coefficient of X1 has dominating effect on Y in the data generating
process.
Generally, bootstrap within method produces lowest RBIAS when true
coefficients of X1 and X2 have equal effect on Y. Furthermore, under random
hot deck imputation, bootstrap across method produces the largest RBIAS
which are far from values produced by the other two estimation methods.
When the common variable used is X2 and X1 has dominating effect on
both Y and Z, resulting RBIAS are larger. This is true for Poisson regression
imputation and MCMC imputation which generally have similar results,
although MCMC imputation has the lowest RBIAS in most cases. In general,
bootstrap within and bootstrap across estimation methods under these two
matching methods have lower RBIAS.
Furthermore, under Poisson regression imputation and MCMC imputation
the differences of RBIAS values among the three estimation procedures are
not large. However, under random hot deck imputation, bootstrap across
method yields large RBIAS values far from values yielded by using Poisson
regression without bootstrap and bootstrap within methods. Bootstrap within
reduces RBIAS in scenario cases when X1 and X2 have equal effect on both Y
and Z while X1 and X2 have low correlation.
Regardless of the common variable, the three estimation procedures yield
similar results although bootstrap within and bootstrap across always have
lower MAE values than those of Poisson regression without bootstrap. For
small sample (size 200), bootstrap within produces lowest MAE values. This is
true for all matching methods. Also, under random hot deck imputation,
bootstrap within always yields lowest MAE values compared to those
produced by using Poisson regression imputation and MCMC imputation.
Moreover, comparable results were produced by using Poisson regression
imputation and MCMC imputation methods, while larger MAE values were
produced by using random hot deck imputation.
Ratio of Data Source A and Data Source B to total sample
When common variable used is X1, RBIAS produced by the three
estimation procedures are comparable when matching methods are Poisson
regression imputation and MCMC imputation. Generally, when X1 has
dominating effect on Y, Poisson regression imputation yields the lowest RBIAS.
Also, lower RBIAS results from using Poisson regression imputation for
scenario cases that satisfy all of the following conditions: (a) X1, X2 have equal

17 | I S I W S C 2 0 1 9