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
- /
- SAS/GRAPH and ODS Graphics
- /
- Funnel Plots - labelling outliers

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

07-27-2016 12:40 PM

Hi all,
I've used Rick Wicklin's post on creating funnel plots (http://blogs.sas.com/content/iml/2011/11/23/funnel-plots-for-proportions.html) successfully on my data. The challenge I'm facing now is figuring out how to label the outliers (above and below) based on either the 95% or 99.5% levels. I can apply labels when the threshold is a set number on the X or Y axis, but I can't seem to find anything on how to do it based on the proportion.
Thanks for your time
Chris
Here's an example of what I'm looking to do, taken from https://blog.ouseful.info/2011/10/31/power-tools-for-aspiring-data-journalists-r/. The article does not provide the code or math to accomplish this, rather it's how to build a simplistic version in R (which doesn't help me).

Has my article or post helped? Please mark as Solution or Like the article!

Accepted Solutions

Solution

07-27-2016
06:59 PM

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

07-27-2016 02:10 PM

You need to merge the data and the limits. To help do that, use all integer values for the limits. You need to make changes in three places:

1) In the IML program, there is a definition of 'n' as

n = T( do(minN, maxN, (maxN-minN)/20) );

Replace that line with all possible trials:

n = T(minN:maxN);

2) Merge the data and the computed limits:

```
/* merge data with proportions */
data Adoptions; merge Adoptions Stats; run;
/* merge control limits */
proc sort data=Adoptions; by Trials; run;
data Funnel;
merge Adoptions(rename=(Trials=N)) Limits;
by N;
Label = Authority;
if L3sd <= Proportion <= U3sd then Label=""; /* delete if w/in +/- 3*sigma */
run;
```

3) Change the SCATTER statement to

scatter x=N y=Proportion / datalabel=Label;

All Replies

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

07-27-2016 12:58 PM

For each observation, you can tell whether it is outside of the 95% confidence interval. Create a label variable that is missing for points inside the CI and .has nonmissing values for points outside the CI, as shown in the article "Label only certain observations with PROC SGPLOT."

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

07-27-2016 01:25 PM

Hi @Rick_SAS

Thanks for the reply - I actually spent quite a bit of time yesterday working thourhg that blog, as I figured it was the easiest thing to do. Where I'm stumped is in the "Funnel" dataset, we have "Authority", "Events", Trials", "Proportion", and "Expected"; then, rows 144 - 164 (in the same dataset) have "N", "L3sd", "L2sd", "U2sd", and "U3sd". If I understand correctly, what I need to do is match the number of events for the individual authority (for example, Halton has n=20) with the closest N (in my case, probably N=26) so that i can get the lower/upper CI of 0.435 and 0.956. Where I'm getting stuck is figuring out how to do this; I tried splitting the tables into 2 and then doing a join in SQL but couldn't get it to work.

Based on the article you linked to, here's my result based on Proportion <=0.5 and Proportion >=0.95. This works fine for the N<=60 or so, but because the Proportions drop (and therefore the CIs get narrower), this does not work for the right side of the graph. What I'd like is all the points above the upper red line and below the lower red line to be labelled.

Thanks again for your help. I'm hoping I have explained this clearly - I'm multi-tasking (trying to do work and this at the same time :-)).

Chris

Has my article or post helped? Please mark as Solution or Like the article!

Solution

07-27-2016
06:59 PM

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

07-27-2016 02:10 PM

You need to merge the data and the limits. To help do that, use all integer values for the limits. You need to make changes in three places:

1) In the IML program, there is a definition of 'n' as

n = T( do(minN, maxN, (maxN-minN)/20) );

Replace that line with all possible trials:

n = T(minN:maxN);

2) Merge the data and the computed limits:

```
/* merge data with proportions */
data Adoptions; merge Adoptions Stats; run;
/* merge control limits */
proc sort data=Adoptions; by Trials; run;
data Funnel;
merge Adoptions(rename=(Trials=N)) Limits;
by N;
Label = Authority;
if L3sd <= Proportion <= U3sd then Label=""; /* delete if w/in +/- 3*sigma */
run;
```

3) Change the SCATTER statement to

scatter x=N y=Proportion / datalabel=Label;

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

07-27-2016 06:59 PM

thanks Rick - that was perfect and I actually think I understand what you did :-)

Have a great evening

Chris

Has my article or post helped? Please mark as Solution or Like the article!

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

07-27-2016 07:09 PM

Although simple, tt's a bit of overkill to have 'n' contain all integers. The important thing is that it contains all the trial values in the data, as well as enough evenly spaced values to visualize the funnel. Then you can do the merge, rather than interpolate.

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

07-27-2016 08:39 PM

thanks for the further explanation. definitely one of the cooler analyses I've worked on in a while!

Chris

Has my article or post helped? Please mark as Solution or Like the article!