peter prop scores
Mark Sanderson-cimino
2023-06-21
Table below shows output from Matchit Package The formula is: group ~ age + sex (as a factor) +PSPRS The “distance variable” is the propensity score. We had to drop ~40 controls because they did not have PSPRSs
print("FULL SAMPLE")## [1] "FULL SAMPLE"summary(m.out.nearest)##
## Call:
## matchit(formula = fmlaMatching, data = df.peter.matchit, method = "nearest",
## link = "logit")
##
## Summary of Balance for All Data:
## Means Treated Means Control Std. Mean Diff.
## distance 0.1592 0.1157 0.5759
## Baseline_Age 66.0000 68.5505 -0.2880
## `as.factor(female0_male1)`0 0.7333 0.5229 0.4758
## `as.factor(female0_male1)`1 0.2667 0.4771 -0.4758
## PSPRSTOT 33.8000 30.1284 0.2417
## Var. Ratio eCDF Mean eCDF Max
## distance 1.2840 0.1772 0.3529
## Baseline_Age 1.2755 0.0815 0.2813
## `as.factor(female0_male1)`0 . 0.2104 0.2104
## `as.factor(female0_male1)`1 . 0.2104 0.2104
## PSPRSTOT 1.2359 0.0645 0.2355
##
## Summary of Balance for Matched Data:
## Means Treated Means Control Std. Mean Diff.
## distance 0.1592 0.1580 0.0159
## Baseline_Age 66.0000 68.0667 -0.2334
## `as.factor(female0_male1)`0 0.7333 0.8000 -0.1508
## `as.factor(female0_male1)`1 0.2667 0.2000 0.1508
## PSPRSTOT 33.8000 35.2000 -0.0922
## Var. Ratio eCDF Mean eCDF Max Std. Pair Dist.
## distance 1.0387 0.0083 0.0667 0.0248
## Baseline_Age 1.4940 0.0688 0.2000 0.7302
## `as.factor(female0_male1)`0 . 0.0667 0.0667 0.7538
## `as.factor(female0_male1)`1 . 0.0667 0.0667 0.7538
## PSPRSTOT 1.2982 0.0575 0.2667 0.8734
##
## Sample Sizes:
## Control Treated
## All 109 15
## Matched 15 15
## Unmatched 94 0
## Discarded 0 0print("CBS SAMPLE")## [1] "CBS SAMPLE"summary(m.out.nearest.CBS)##
## Call:
## matchit(formula = fmlaMatching, data = df.peter.matchit.CBS,
## method = "nearest", link = "logit")
##
## Summary of Balance for All Data:
## Means Treated Means Control Std. Mean Diff.
## distance 0.1527 0.1367 0.3652
## Baseline_Age 64.8000 67.2581 -0.2310
## `as.factor(female0_male1)`0 0.6000 0.4839 0.2370
## `as.factor(female0_male1)`1 0.4000 0.5161 -0.2370
## PSPRSTOT 25.6000 23.9355 0.1742
## Var. Ratio eCDF Mean eCDF Max
## distance 0.8457 0.1294 0.3419
## Baseline_Age 1.2541 0.1399 0.4129
## `as.factor(female0_male1)`0 . 0.1161 0.1161
## `as.factor(female0_male1)`1 . 0.1161 0.1161
## PSPRSTOT 0.6966 0.0854 0.2129
##
## Summary of Balance for Matched Data:
## Means Treated Means Control Std. Mean Diff.
## distance 0.1527 0.1556 -0.0680
## Baseline_Age 64.8000 64.4000 0.0376
## `as.factor(female0_male1)`0 0.6000 0.6000 0.0000
## `as.factor(female0_male1)`1 0.4000 0.4000 0.0000
## PSPRSTOT 25.6000 27.0000 -0.1465
## Var. Ratio eCDF Mean eCDF Max Std. Pair Dist.
## distance 0.9750 0.0278 0.2 0.0682
## Baseline_Age 7.6486 0.1545 0.4 0.8271
## `as.factor(female0_male1)`0 . 0.0000 0.0 0.4000
## `as.factor(female0_male1)`1 . 0.0000 0.0 0.4000
## PSPRSTOT 0.3609 0.1200 0.4 1.1512
##
## Sample Sizes:
## Control Treated
## All 31 5
## Matched 5 5
## Unmatched 26 0
## Discarded 0 0print("PSP SAMPLE")## [1] "PSP SAMPLE"summary(m.out.nearest.PSP)##
## Call:
## matchit(formula = fmlaMatching, data = df.peter.matchit.PSP,
## method = "nearest", link = "logit")
##
## Summary of Balance for All Data:
## Means Treated Means Control Std. Mean Diff.
## distance 0.1741 0.1059 0.6769
## Baseline_Age 66.6000 69.0641 -0.2935
## `as.factor(female0_male1)`0 0.8000 0.5385 0.6538
## `as.factor(female0_male1)`1 0.2000 0.4615 -0.6538
## PSPRSTOT 37.9000 32.5897 0.3279
## Var. Ratio eCDF Mean eCDF Max
## distance 1.4917 0.2164 0.4231
## Baseline_Age 1.4059 0.0954 0.2154
## `as.factor(female0_male1)`0 . 0.2615 0.2615
## `as.factor(female0_male1)`1 . 0.2615 0.2615
## PSPRSTOT 1.3862 0.0976 0.3128
##
## Summary of Balance for Matched Data:
## Means Treated Means Control Std. Mean Diff.
## distance 0.1741 0.1772 -0.0309
## Baseline_Age 66.6000 69.1000 -0.2978
## `as.factor(female0_male1)`0 0.8000 1.0000 -0.5000
## `as.factor(female0_male1)`1 0.2000 0.0000 0.5000
## PSPRSTOT 37.9000 35.8000 0.1297
## Var. Ratio eCDF Mean eCDF Max Std. Pair Dist.
## distance 0.9095 0.0116 0.1 0.0836
## Baseline_Age 1.7291 0.1115 0.3 0.8933
## `as.factor(female0_male1)`0 . 0.2000 0.2 0.5000
## `as.factor(female0_male1)`1 . 0.2000 0.2 0.5000
## PSPRSTOT 1.4347 0.0804 0.2 0.6606
##
## Sample Sizes:
## Control Treated
## All 78 10
## Matched 10 10
## Unmatched 68 0
## Discarded 0 0Comparison of propensity scores between groups. There was orginally an outlier in the treatment group, this participant was removed based on our previous convo. The age of the participant was 44.
print("FULL SAMPLE")## [1] "FULL SAMPLE"plot(m.out.nearest, type = "jitter", interactive = FALSE)
print("CBS SAMPLE")## [1] "CBS SAMPLE"plot(m.out.nearest.CBS, type = "jitter", interactive = FALSE)
print("PSP SAMPLE")## [1] "PSP SAMPLE"plot(m.out.nearest.PSP, type = "jitter", interactive = FALSE)
Density Plots Black line = cases, gray line = controls Left graph is full dataset, right graph is matched dataset To me the matching is better, but it’s not perfect.
fmlaMatching<-NULL
fmlaMatching <- as.formula("Rockit_1_4RNTNI_0~Baseline_Age+female0_male1+PSPRSTOT")
m.out.nearest <- matchit(fmlaMatching, method = "nearest",
data = df.peter.matchit, link = "logit")
m.out.nearest.CBS <- matchit(fmlaMatching, method = "nearest",
data = df.peter.matchit.CBS, link = "logit")
m.out.nearest.PSP <- matchit(fmlaMatching, method = "nearest",
data = df.peter.matchit.PSP, link = "logit")
print("FULL SAMPLE")## [1] "FULL SAMPLE"print(plot(m.out.nearest, type = "density", interactive = FALSE,
which.xs = ~Baseline_Age + female0_male1 + PSPRSTOT))
## A matchit object
## - method: 1:1 nearest neighbor matching without replacement
## - distance: Propensity score
## - estimated with logistic regression
## - number of obs.: 124 (original), 30 (matched)
## - target estimand: ATT
## - covariates: Baseline_Age, female0_male1, PSPRSTOTprint("CBS SAMPLE")## [1] "CBS SAMPLE"print(plot(m.out.nearest.CBS, type = "density", interactive = FALSE,
which.xs = ~Baseline_Age + female0_male1 + PSPRSTOT))
## A matchit object
## - method: 1:1 nearest neighbor matching without replacement
## - distance: Propensity score
## - estimated with logistic regression
## - number of obs.: 36 (original), 10 (matched)
## - target estimand: ATT
## - covariates: Baseline_Age, female0_male1, PSPRSTOTprint("PSP SAMPLE")## [1] "PSP SAMPLE"print(plot(m.out.nearest.PSP, type = "density", interactive = FALSE,
which.xs = ~Baseline_Age + female0_male1 + PSPRSTOT))
## A matchit object
## - method: 1:1 nearest neighbor matching without replacement
## - distance: Propensity score
## - estimated with logistic regression
## - number of obs.: 88 (original), 20 (matched)
## - target estimand: ATT
## - covariates: Baseline_Age, female0_male1, PSPRSTOT