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
- /
- Stat Procs
- /
- extreme ORs and procudre for continuous independen...

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

a month ago

I have a categorical dependent variable and a continuous independent variable. Which procedure is best suited to find significance between the two. The Proc Reg gives one t-value for all levels of the dependent categorical variable. And it doesn't even provide the OR for the same. What other procedure can be used in order to find the OR for this procedure?

Also, I used Generalized Logit function for finding out the relation between categorical dependent and independent variables for which proportional odds was NOT significant. But I received extreme values such as the follows. OR= (>999.999 ). 95% CI ( <0.001, >999.999). How can such error be rectified or will it be right to use this value?

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

a month ago

This sounds like a logistic multinomial regression. Are the categorical levels ordinal or nominal?

See the example here for nomical logistic regression and interpreting the output. If you have ORs that are 0 or 999.99 it means you don't have data for all your categories and it really can't come up with an estimate.

It sounds like you only have 2 variables, so I assumed you started out your analysis with an ANOVA to check the differences between the groups or such?

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

a month ago

For first part, where I used proc REG, the independent variable is continuous and the dependent variable is ordinal. REG gave only one t-value. Which procedure would be suitable? glogit is not apt for it will give as many values of the OR as many continuous variables are present.

For the second question too, I had ordinal independent variable and ordinal dependent variable. In case every category doesn't have enough values, should i report the ORs for another category? like instead of 1vs2 , I could report 1vs 3.

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

a month ago

I think you need glogit for the dependent variable. I thought the OR for continuous variable would be one value that represents one unit increase. Based on the very little you've said here I still think Proc Logistic is correct but you may want to wait on some more answers. You may want to consider posting a question, with more details, in the statistical procedure forum instead of this generic forum.

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

a month ago

Guess I'll have to post again with more information

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

a month ago - last edited a month ago

Reeza, I have a lot of variables in the dataset and only one ordinal dependent variable. I'm finding the association between this outcome variable and one independent variable at a time. I did not use ANOVA since I was supposed to find the association of these independent variables with an ordinal response variable. So I opted to use Glogit and Ordinal logit procedure.

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

4 weeks ago

See my response in your other posting in this forum about using model selection in PROC LOGISTIC for an ordinal response allowing for nonproportional odds.