Another Little Gem

Here's another great little free program, G*Power 3. It does many of the common anova power analyses but also includes manova and Hotelling's T-squared. Throw in multiple regression and you have a pretty useful package.

There are versions for both windows and macs. This in itself is pretty unusual.

The authors are Franz Faul, Edgar Erdfelder and Axel Buchner. They all appear to be psychologists but that is no reason to avoid G*Power. I highly recommend it. The website is at Heinrich Heine University Dusseldorf and here is the link.

pbe

What The Heck(man) Is Going On?

We had a client come in with a question about a Heckman selection model that was giving her trouble. She had run it several weeks earlier and everything was working fine. She had some missing data among her predictors and decided to do a multiple imputation. After imputing the missing data she got the following error message:

     Dependent variable never censored due to selection.

She couldn't figure out what was wrong until I asked her if she had also imputed the response (dependent) variable. Instantly, she realized what the problem was. Since she had imputed all the variables in her dataset, there were no longer any missing values on her response variable and therefore no way to estimate a selection model.

pbe

Parallel Universe

I am a Mac user. This creates problems in that there are several stat packages that only run on Windows. For the past couple of years the solution has been for me to Timbuktu into a campus computer to use Windows based software. This solution is relatively slow and clunky.

That was then, this is now. I have installed Parallels Desktop on my relatively new MacBook Pro. Then, after a few hours installing Windows and a few more hours installing stat software (I had problems installing SAS), I'm all set to go. I now run Mplus and SAS on my Mac. And actually, they run pretty fast as long as I close down some of the larger programs on the Mac side. Overall, my throughput is much faster than when I was using Timbuktu. So, for now, I am a happy camper working in a parallel universe.

pbe

The Parent Trap

Here is a question emailed to us by a client:

I have 4 variables from which I want to create a new variable.

v1 = do you live with mom? (0, 1)
v2 = do you live with dad?
v3 = do you live with stepmother?
v4 = do you live with stepfather?

I want to create a variable PCOMP and code
1 if live with mom and dad
2 if live with mom and stepfather
3 if live with stepmother and dad
4 if live with mom only
5 if live with dad only
6 Other

I'd also like to identify who is from a two-parent household.

How do I do this?

There are many ways this could be handled, here is the one I suggested:

pcomp = v1 + 10*v2 + 100*v3 + 1000*v4

recode pcomp 11=1 1001=2 110=3 1=4 10=5 else=6

two_parent = pcomp < 4

Note: This is in pseudo-code and can be adapted to any stat package that supports recode. Also, the client will have to decide how to deal with missing values as this approach will not work with missing values for v1-v4.

pbe

Power Series

Today we had the second in our series of three power analysis presentations. Jason Cole (UCLA alum and outside consultant) talked in depth about effect size, the proper alpha level and multiplicity issues to an audience of 42. Within the next week, we hope to post the audio portion of Jason's presentation along with a pdf of his powerpoint slides.

Last week, for Christine's presentation, we had almost sixty people in attendance. This created problems because the Visualization Portal is not meant to hold that many people. For this week's presentation we had to have people register on-line to limit the number attending.

We are looking forward to another big crowd for our third power analysis presentation in two weeks.

pbe

A Rose By Any Other Name...

UCLA Academic Technology Services Statistical Consulting is pleased to announce that Rose Medeiros starts work today as our new full-time stat consultant. She comes to us from the University of New Hampshire where she is completing her PhD in sociology. She just spent the last five days driving cross country with her mother and her one-eyed cat named Bentley.

Rose has experience using SPSS, Stata, HLM, Mplus and sometimes R. She "enjoys" multilevel modeling, longitudinal data analysis and structural equation modeling.

And, of course, Rose is "famous" for the Stata ado-program -njc- that provides timely quotes from the statistically-minded geographer Nicholas J Cox (note: the -njc- program does not use maximum likelihood estimation).

Please drop by walk-in consulting and say "Hi," to Rose.

pbe

A Little Gem of a Program

A tip by legendary stat consultant Michael Mitchell pointed us to a little gem of a program, Optimal Design.

Optimal Design does power analysis for a wide variety of multilevel level and repeated measures designs. The program is a product of the Survey Research Center of the Institute of Social Research at the University of Michigan. The program is fast, easy to use, and the best part, it's free. You can find out about it at http://sitemaker.umich.edu/group-based/optimal_design_software.

This program comes at an opportune time for us since ATS Stat Consulting will be giving the first of three presentations on power analysis starting next Wednesday, February 14, 2007 at 10am. The other two presentations will be Feb 14th and Mar 7th.

pbe