turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- Base SAS Programming
- /
- Missing observations in Multivariate Regression

Topic Options

- RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

12-02-2017 07:52 PM

Dear all,

I am running a multivariate logistic regression model assessing for occurrence of dependant event X (0-didn't occur 1-occurred) with about 200,00 weighted observations using survey data.

I have multiple independant variables, (both continuous and a few categorical).

Many of the independant variables are missing in about 5% of the data observations (the same 5% of observations are missing data).

Will this make my conclusion inaccurate? I am not sure of how this will effect the overall model, and how it will effect the variables that do not have missing information.

Please clarify if I can run the analysis, or will it cause a major issue. I would prefer to include those 5% of observations if possible, because they have information for some variables.

Thank You

Accepted Solutions

Solution

12-07-2017
03:36 PM

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to sasnewbie12

12-02-2017 08:27 PM - edited 12-02-2017 08:28 PM

Hi,

By default SAS will exclude observation if it has even a single variable with missing value. Model excluding 5% missing observations will not make a significant shift in conclusions if problem is related to draw inferences only. If your project is related to predictive modeling that requires scoring new data sets then you need to impute missing variables. Please look into proc stdize that provides various imputation options.

All Replies

Solution

12-07-2017
03:36 PM

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to sasnewbie12

12-02-2017 08:27 PM - edited 12-02-2017 08:28 PM

Hi,

By default SAS will exclude observation if it has even a single variable with missing value. Model excluding 5% missing observations will not make a significant shift in conclusions if problem is related to draw inferences only. If your project is related to predictive modeling that requires scoring new data sets then you need to impute missing variables. Please look into proc stdize that provides various imputation options.

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to stat_sas

12-03-2017 10:41 AM

Just want to make sure, will observations that are missing a value for a variable that is not included in your model as a independant or dependant variable also be excluded?

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to sasnewbie12

12-03-2017 03:44 PM

No. Only for the variables included in your model. For example, this regression

```
data class;
set sashelp.class;
if _n_<5 then age=.;
run;
proc glm data=class;
model height=weight;
run;quit;
```

uses all 19 observations in the data set even though age contains missing values for 5 observations. But since age is not included in your model, the observations are not excluded.

This regression however

```
data class1;
set sashelp.class;
if _n_<5 then weight=.;
run;
proc glm data=class1;
model height=weight;
run;quit;
```

uses only 15 observations because weight contains missing values for 5 observations and weight is included in your model.

- Mark as New
- Bookmark
- Subscribe
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to sasnewbie12

12-03-2017 12:04 AM

You might find this paper interesting:

http://support.sas.com/resources/papers/proceedings16/SAS3520-2016.pdf

PG