How can I isolate influencial obsevations in logistic regression outliers?

[maroulator]

To whom it may concern,

I am running the following snippet of code and I am trying to capture outliers using the Pearson Residual, Hat Matrix Diagonal, and DfBeta values. This a very straightforward task if my dataset is less than 10,000 observations. My problem, however, is that once I start working on a dataset with observations in the millions, sifting through my output file becomes an extremely cumbersome process; this is because my output file lists out every single one of the observations in the dataset along with the associated leverage/influence metrics and graphs. Is anyone aware of a way to have my output only display the observations that have a DfBeta, or a Person Residual, or a Hat Matrix diagonal value above a certain level?

proc logistic data=Dataset descending;

model Y=var1 var2 var3 var4/ plcl plrl waldcl waldrl

                                                lackfit rsq

                                                influence iplots

                                                itprint;

run;

[Reeza · Accepted solution]

Put them into a data set instead and then filter as you would any other source.

Check the OUTPUT statement and the options available to capture the data.

---

@maroulator wrote:

To whom it may concern,

I am running the following snippet of code and I am trying to capture outliers using the Pearson Residual, Hat Matrix Diagonal, and DfBeta values. This a very straightforward task if my dataset is less than 10,000 observations. My problem, however, is that once I start working on a dataset with observations in the millions, sifting through my output file becomes an extremely cumbersome process; this is because my output file lists out every single one of the observations in the dataset along with the associated leverage/influence metrics and graphs. Is anyone aware of a way to have my output only display the observations that have a DfBeta, or a Person Residual, or a Hat Matrix diagonal value above a certain level?

proc logistic data=Dataset descending;

model Y=var1 var2 var3 var4/ plcl plrl waldcl waldrl

                                                lackfit rsq

                                                influence iplots

                                                itprint;

run;

---

[ballardw]

Adding ods output to the body of the Proc logistic code:

proc logistic data=Dataset descending;
model Y=var1 var2 var3 var4/ plcl plrl waldcl waldrl
                           lackfit rsq
                           influence iplots
                           itprint;
ods output influence=myinfluencedatasetname;
run;

will place the content of the influence results table into a dataset.

Filter or sort as interested.