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
- /
- Analytics
- /
- Data Mining
- /
- Using proc logistic in SAS

Topic Options

- Subscribe to 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
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

04-19-2016 04:42 PM

Hi,

I am using proc logistic in order to create a logistic regression to predict a var called res1 base in some variables called var1,var2...., var8 ( 8 vars)

proc logistic data=test outmodel=modelo plots=ROC ;

class var1 var2 ............... var8;

model res1 = var1 var2 ............... var8 /* selection=forward */;

quit;

I would like to know which is the best way to notice which are the most important variables in my modelo, the vars that have more prediction capacity. Prior to execute the proc logistic I do a "proc discrim" in order to get the most important vars, but I don't know if is the best way...

Also I would like to know if there is anyway to gest the most important vars for the model, the most importance vars for proc logistics. For example in R y do importance(name_of_model)

Any advice will be greatly apreciatted

Accepted Solutions

Solution

04-25-2016
05:02 AM

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

04-20-2016 04:26 AM

"My question is...can I force procc logistic to get only one parameter for variable??"

Sorry. I can't understand what is your mean. If that variable was category variable, maybe you could check EFFECT statement.

Something like. maybe you could try other combination.

```
proc logistic data=sashelp.class;
class age;
effect new_age=collection(age);
model sex(ref='M')=new_age weight height/clodds=wald;
run;
```

"What output coeficients are good to validate the model?"

About C, as far as I know it is the area under ROC curve. which means it is bigger ,the model is better .

All Replies

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

04-19-2016 09:20 PM - edited 04-19-2016 09:31 PM

1)You can check ODDS RATIO of each variable, the most of it has the most power .

2)selection=stepwise , the first one enter the model should be the most important ? Just guess.

E.X.

```
proc logistic data=sashelp.class;
model sex(ref='M')=age weight height/clodds=both;
run;
```

Seems AGE is the most import variable.

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

04-20-2016 04:11 AM

Thanks for your help.

I have for one variable different parameters in the model depending on the range of value of the variable. That is:

My question is...can I force procc logistic to get only one parameter for variable??

Other question:

What output coeficients are good to validate the model?

Thanks in advance?

Thanks

Solution

04-25-2016
05:02 AM

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

04-20-2016 04:26 AM

"My question is...can I force procc logistic to get only one parameter for variable??"

Sorry. I can't understand what is your mean. If that variable was category variable, maybe you could check EFFECT statement.

Something like. maybe you could try other combination.

```
proc logistic data=sashelp.class;
class age;
effect new_age=collection(age);
model sex(ref='M')=new_age weight height/clodds=wald;
run;
```

"What output coeficients are good to validate the model?"

About C, as far as I know it is the area under ROC curve. which means it is bigger ,the model is better .

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

04-22-2016 04:41 AM

In the below screen your C statistic is very high and ideally it should be more than 0.5 for the model to be fit.

Also the concordarnce statistics is robust at 97+ which shows the probability of events where discordance is non event. so in your case you have a very high probability of events.

To check which var are significant or insignificant check the parameter estimate of each independent var & its corresponding p value as well.

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

04-22-2016 04:54 AM

Thanks very much...., very useful

Do you know how to interpret this coeffs?

AIC, SC , -2LogL

Thanks

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

04-22-2016 05:21 AM

That is model goodness-fit statistic. The lower imply the model is better.