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
- /
- How can I assess the model through the diagnostics...

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

11-22-2015 05:18 PM - edited 11-23-2015 09:16 AM

The simple linear regression model is E(Yi) *=*α+β Xi

When the regression is done, I get the diagnostics panel as below. I was required to access the model quality through the diagnostics panel and I also see some explanation through SAS support:

but It is still not very understandable. I hope someone can expain to me through the following example. Thanks in advance.

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

11-22-2015 09:08 PM

I think you mean assess, not access.

One of the assumptions for linear regression is:

Errors are normally distributed.

If you look at the histogram in the bottom right corner (plot7) you can see that it's not quite normally distributed so you may have some violations.

Cooks Distance (plot6) - used to detect outliers

https://en.wikipedia.org/wiki/Cook%27s_distance

You can see some large Cook's D values so there are likely some outliers. It would be worth investigating those observations.

Residual Quantile Plot (Plot4) - look for a straight diagonal line.

http://onlinestatbook.com/2/advanced_graphs/q-q_plots.html

You have some stragglers, so most likely those outliers.

Basically take each plot, google it and see what you should be looking for and does your data match up. If not, why.

One of the assumptions for linear regression is:

Errors are normally distributed.

If you look at the histogram in the bottom right corner (plot7) you can see that it's not quite normally distributed so you may have some violations.

Cooks Distance (plot6) - used to detect outliers

https://en.wikipedia.org/wiki/Cook%27s_distance

You can see some large Cook's D values so there are likely some outliers. It would be worth investigating those observations.

Residual Quantile Plot (Plot4) - look for a straight diagonal line.

http://onlinestatbook.com/2/advanced_graphs/q-q_plots.html

You have some stragglers, so most likely those outliers.

Basically take each plot, google it and see what you should be looking for and does your data match up. If not, why.

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

11-23-2015 07:00 AM

The Link you have provided it almost tells every thing.In your case the Rsqure is is only .40 that means there are other variables which are accounted for .60 or 60%. Your Adj Rsquare is even more less 37% . Find out all those variables and do multiple regression . Your plots (No1)looks dont look like a fan shape or semicircle that which menas it is ok . You have one outlier as per the C'D panel (No 6).

Sayng so , please have a look at your data , Remove the outlier.Please understand with the theory first. I am also just learning ,but not reached Regression Analysis yet .so i could not provide much ,may be others can help.

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

11-23-2015 09:30 AM

Thanks pearsoninst, it is very nice of you.

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

11-23-2015 09:25 AM

Thanks Reeza, it helps.