0

I performed propensity score matching using the nearest neighbor method. To assess the balance by matching I used the bal.tab function of the COBALT package setting the thresholds for mean differences to ".05".

# Balance tables with thresholds for mean differences and means pre and post matching
    bal.tab(m.outtest, continuous = "std", binary = "raw", disp = c("means"), un = TRUE, stats = c("m"),thresholds = c(m = .05))

I received a result about balance measure,but I did not find any information on which statistical tests the p-values are based on, which result in "balanced" or "not balanced".

enter image description here

Call

matchit(formula = icu_mort ~ age + sex + duration_h, data = testdata1, 
    method = "nearest", distance = "glm", replace = FALSE, caliper = 0.2, 
    std.caliper = FALSE, ratio = 1)

Balance Measures
               Type   M.0.Un   M.1.Un Diff.Un  M.0.Adj  M.1.Adj Diff.Adj     M.Threshold
distance   Distance   0.2153   0.2627  0.5343   0.2615   0.2627   0.0132     Balanced, <0.05
age         Contin.  59.8966  67.4071  0.5220  67.6637  67.4071  -0.0178     Balanced, <0.05
sex_w        Binary   0.4419   0.3628 -0.0790   0.3717   0.3628  -0.0088     Balanced, <0.05.  
duration_h  Contin. 348.9122 367.6069  0.0496 338.4327 367.6069   0.0774 Not Balanced, >0.05

Balance tally for mean differences
                    count
Balanced, <0.05         3
Not Balanced, >0.05     1

Variable with the greatest mean difference
   Variable Diff.Adj         M.Threshold
 duration_h   0.0774 Not Balanced, >0.05

Usually a p-value >0.05 means that there are no significant differences between the groups, but this is what is called "not balanced" here, that means there are differences between treatment group and control group after matching.

enter image description here

Could someone please explain to me this (apparent) inconsistency regarding the meaning of the p-values and on which statistical test the p-values are based?

neilfws
  • 32,751
  • 5
  • 50
  • 63
user19939387
  • 1
  • 1
    This post is not really a good fit for Stack Overflow as it is not a coding question. You might want to ask on [Cross Validated](https://stats.stackexchange.com/), or if it is not in the docs for this particular package ask on [their github issues](https://github.com/ngreifer/cobalt/issues). I have voted for this to be migrated to Cross Validated. – SamR Sep 07 '22 at 10:30
  • Greetings! Usually it is helpful to provide a minimally reproducible dataset for questions here. One way of doing this is by using the `dput` function. You can find out how to use it here: https://youtu.be/3EID3P1oisg – Shawn Hemelstrand Sep 10 '22 at 06:21
  • Thanks for the hint about their github issues! – user19939387 Sep 13 '22 at 07:48
  • You will find the answer to my question on https://github.com/ngreifer/cobalt/issues/62 – user19939387 Sep 14 '22 at 06:50

0 Answers0