I'm running many regressions and am only interested in the effect on the coefficient and p-value of one particular variable. So, in my script, I'd like to be able to just extract the p-value from the glm summary (getting the coefficient itself is easy). The only way I know of to view the p-value is using summary(myReg). Is there some other way?
e.g.:
fit <- glm(y ~ x1 + x2, myData)
x1Coeff <- fit$coefficients[2] # only returns coefficient, of course
x1pValue <- ???
I've tried treating fit$coefficients as a matrix, but am still unable to simply extract the p-value.
Is it possible to do this?
Thanks!
You want
coef(summary(fit))[,4]
which extracts the column vector of p values from the tabular output shown by summary(fit). The p-values aren't actually computed until you run summary() on the model fit.
By the way, use extractor functions rather than delve into objects if you can:
fit$coefficients[2]
should be
coef(fit)[2]
If there aren't extractor functions, str() is your friend. It allows you to look at the structure of any object, which allows you to see what the object contains and how to extract it:
summ <- summary(fit)
> str(summ, max = 1)
List of 17
$ call : language glm(formula = counts ~ outcome + treatment, family = poisson())
$ terms :Classes 'terms', 'formula' length 3 counts ~ outcome + treatment
.. ..- attr(*, "variables")= language list(counts, outcome, treatment)
.. ..- attr(*, "factors")= int [1:3, 1:2] 0 1 0 0 0 1
.. .. ..- attr(*, "dimnames")=List of 2
.. ..- attr(*, "term.labels")= chr [1:2] "outcome" "treatment"
.. ..- attr(*, "order")= int [1:2] 1 1
.. ..- attr(*, "intercept")= int 1
.. ..- attr(*, "response")= int 1
.. ..- attr(*, ".Environment")=<environment: R_GlobalEnv>
.. ..- attr(*, "predvars")= language list(counts, outcome, treatment)
.. ..- attr(*, "dataClasses")= Named chr [1:3] "numeric" "factor" "factor"
.. .. ..- attr(*, "names")= chr [1:3] "counts" "outcome" "treatment"
$ family :List of 12
..- attr(*, "class")= chr "family"
$ deviance : num 5.13
$ aic : num 56.8
$ contrasts :List of 2
$ df.residual : int 4
$ null.deviance : num 10.6
$ df.null : int 8
$ iter : int 4
$ deviance.resid: Named num [1:9] -0.671 0.963 -0.17 -0.22 -0.956 ...
..- attr(*, "names")= chr [1:9] "1" "2" "3" "4" ...
$ coefficients : num [1:5, 1:4] 3.04 -4.54e-01 -2.93e-01 1.34e-15 1.42e-15 ...
..- attr(*, "dimnames")=List of 2
$ aliased : Named logi [1:5] FALSE FALSE FALSE FALSE FALSE
..- attr(*, "names")= chr [1:5] "(Intercept)" "outcome2" "outcome3" "treatment2" ...
$ dispersion : num 1
$ df : int [1:3] 5 4 5
$ cov.unscaled : num [1:5, 1:5] 0.0292 -0.0159 -0.0159 -0.02 -0.02 ...
..- attr(*, "dimnames")=List of 2
$ cov.scaled : num [1:5, 1:5] 0.0292 -0.0159 -0.0159 -0.02 -0.02 ...
..- attr(*, "dimnames")=List of 2
- attr(*, "class")= chr "summary.glm"
Hence we note the coefficients component which we can extract using coef(), but other components don't have extractors, like null.deviance, which you can extract as summ$null.deviance.
I've used this technique in the past to pull out predictor data from summary or from a fitted model object:
coef(summary(m))[grepl("var_i_want$",row.names(coef(summary(m)))), 4]
which lets me easily edit which variable I want to get data on.
Or as pointed out be @Ben, use match or %in%, somewhat cleaner than grepl:
coef(summary(m))[row.names(coef(summary(m))) %in% "var_i_want" , 4]
Instead of the number you can put directly the name
coef(summary(fit))[,'Pr(>|z|)']
the other ones available from the coefficient summary:
Estimate Std. Error z value Pr(>|z|)
Well, this would be another way, however not the most efficient way to perform it :
a = coeftable(model).cols[4]
pVals = [ a[i].v for i in 1:length(a) ]
This ensures that the extracted values from the glm is not in StatsBase. Therein, you can play around with pVals as per your requirement. Hope it helps, Ebby
来源：https://stackoverflow.com/questions/23838937/extract-pvalue-from-glm