CEM: Stata vs R

[Martine Blom]

Dear StataListers,

I can't get why the analyzes below, performed in Stata / R, do not produce the same result.

Please provide your suggestions for the explanation for these differences.

Thanks,
Martine

PS (how) can I display the actual cutpoints in Stata (imb)?


-- STATA
. import delimited "G:\Recidive\Projecten\Verkeer\Algemeen\Artike l effectiviteit ASP tijdens\Data\Data voor R.csv", clear
(8 vars, 8,627 obs)

. imb d_sekse lftbegrec d_addinfo_ind_best lftinsz1 vgalgexuz vgverkexuz vgrijoiuz, treatment (treat)
(using the scott break method for L1 distance)

Multivariate L1 distance: .68243298

Univariate imbalance:
L1 mean min 25% 50% 75% max
d_sekse .01557 -.01557 0 0 0 0 0
lftbegrec .0687 -1.5626 0 -1 -2 -1 -5
d_addinfo_ind_best .00497 -.00497 0 0 0 0 0
lftinsz1 .08852 .91045 0 1 1 2 -12
vgalgexuz .14912 -1.2611 0 0 -1 -1 78
vgverkexuz .09852 -.41727 0 0 0 -1 4
vgrijoiuz .11859 -.36406 0 0 0 0 -4


--R
> data <- read.csv(file = "G:\\Recidive\\Projecten\\Verkeer\\Algemeen\\A rtik el effectiviteit ASP tijdens\\Data\\Data voor R.csv")
> cov <- c("D_SEKSE", "LFTBEGREC", "D_ADDINFO_IND_BEST","LFTINSZ1", "VGALGEXUZ", "VGVERKEXUZ", "VGRIJOIUZ")
> imb <- imbalance(group = data$TREAT, data = data[cov])
> imb

Multivariate Imbalance Measure: L1=0.665
Percentage of local common support: LCS=11.2%

Univariate Imbalance Measures:

statistic type L1 min 25% 50% 75% max
D_SEKSE -0.015571508 (diff) 0.015571508 0 0 0 0 0
LFTBEGREC -1.562561051 (diff) 0.025019213 0 -1 -2 -1 -5
D_ADDINFO_IND_BEST -0.004970458 (diff) 0.004970458 0 0 0 0 0
LFTINSZ1 0.910454843 (diff) 0.045867098 0 1 1 2 -12
VGALGEXUZ -1.261056065 (diff) 0.146272471 0 0 -1 -1 78
VGVERKEXUZ -0.417271627 (diff) 0.097957912 0 0 0 -1 4
VGRIJOIUZ -0.364063976 (diff) 0.118028583 0 0 0 0 -4

[Scott Merryman]

Aren't they equal? cem in Stata report L1 and then the mean difference. cem in R reports the difference and then L1 statistics.

In Stata the -cem- option -showbreaks- displays the cutpoints used for each variable