Table 1 : Monte carlo simulation data.


    n=600 n=200 n=50

Effect True
Value
NB
Coverage
(True Value)
NB
Coverage (Non-Zero)
OLS Coverage
(Non-Zero)
NB
Coverage
(True Value)
NB
Coverage
(Non-Zero)
OLS
Coverage
(Non-Zero)
NB
Coverage
(True Value)
NB
Coverage
(Non-Zero)
OLS
Coverage
(Non-Zero)

Shift 0.44 0.948 0.986 0.982 0.947 0.609 0.580 0.926 0.211 0.162
Difficulty -0.58 0.938 1.000 1.000 0.946 0.818 0.809 0.926 0.334 0.302
Dehydrated -0.57 0.960 0.994 0.993 0.951 0.687 0.694 0.934 0.319 0.308
Coop 0.44 0.952 0.966 0.961 0.936 0.618 0.602 0.928 0.220 0.192
Dispersion 0.56 0.954 -- -- 0.938 -- -- 0.895 -- --
AIC NB -- -- 1559 -- -- 522 -- -- 136 --
AIC OLS -- -- 1990 -- -- 659 -- -- 159 --
Difference -- -- 431 -- -- 137 -- -- 23 --
Lower 95% -- -- 427 -- -- 135 -- -- 22.6 --
Upper 95% -- -- 434 -- -- 139 -- -- 24.1 --
Percent -- -- 21.6% -- -- 20.6% -- -- 14.3% --

Effect   Diff in Adj
Means
(OLS-NB)
Lower
95%
Upper
95%
Diff in Adj
Means
(OLS-NB)
Lower
95%
Upper
95%
Diff in Adj
Means
(OLS-NB)
Lower
95%
Upper
95%

Shift (Day) -- 0.146 0.144 0.149 0.156 0.151 0.161 0.197 0.186 0.207
Shift
(Night)
-- 0.010 0.007 0.012 0.025 0.021 0.029 0.076 0.065 0.088
Difficult
(No)
-- -0.003 -0.006 0.001 0.015 0.009 0.020 0.060 0.046 0.075
Difficult
(Yes)
-- 0.116 0.114 0.118 0.124 0.120 0.128 0.163 0.154 0.171
Dehydrated
(No/Unk)
-- 0.018 0.015 0.022 0.034 0.028 0.040 0.068 0.051 0.085
Dehydrated
(Yes)
-- 0.095 0.093 0.097 0.104 0.101 0.107 0.139 0.132 0.146
Cooperative
Child (No)
-- 0.130 0.128 0.133 0.140 0.136 0.144 0.191 0.182 0.200
Cooperative
Child (Yes)
-- 0.027 0.024 0.030 0.039 0.033 0.045 0.079 0.063 0.095

The above data are the output from Monte Carlo simulations that were done to confirm the validity of different model assumptions. Simulations were run for multiple sample
sizes. The table lists the true values of the parameters first. The initials NB stand for 'Negative Binomial,' and the initials 'OLS' stand for ordinary least squares. The coverage estimates
consists of examination of 95% confidence intervals for either the true parameter value or zero (as appropriate). Also shown are AIC information for each model.

Brinkley et al.Journal of Medical Statistics and Informatics  2014 2:6DOI : 10.7243/2053-7662-2-6