Red-R

Red-R seems to have some interesting ideas behind it. Of course it needs some time to mature.

Another tablet, looking good

Based on this post, Motorola is going to release an Android-based tablet in the coming fall.

Pdf support in ebook reader

Based on this review, the Sony reader (daily edition) has better pdf support than both nook and kindle, and should be the #1 choice for handling pdf books (now costs $249). Here is a more detailed explanation of how sony reader deals with pdf files.

This web site deserves some attention, and this is an excellent review.

Recent emmigration from China...

This is an interesting article (in Chinese).

Ebook reader price cut

All major ebook readers have recently experienced major price cut (for example, Kindle DX, Nook). It seems that the Apple Ipad has something to do with it.

Looking forward to HP's webOS-based slate computer

From what I have read, webOS is a really cool and mature Linux-based OS for handhold device. Since HP is planning to release a webOS-based slate computer ("palmpad"?) later this year, I think it will be worth waiting for.

Living in China

This article (in Chinese) is really interesting...

Flexmix taks long time...

I am running a mixture logit model on a sample of 10,000 cases. The model takes a long time (3-4 hours) to run, which makes it nearly impossible to try different specifications and such. It probably makes sense to use a subsample for exploratory analysis then fit the final model on the full sample.

This post describes how to draw a random sample in R.

Npmlreg and Flexmix

In my attempt to estimate a mixture regression model for the long-term impact of prenatal famine exposure, I compared results obtained from NPMLREG and Flexmix using the "rainfall" data set came with the package "forward" and was able to achieve identical results from:

m.np <- allvc(cbind(Cases, Total-Cases) ~ Rain, random= ~1|ID, data=rainfall, k=2, family=binomial(logit))

and 

m.mix <- stepFlexmix(cbind(Cases, Total-Cases) ~ 1|ID, model=FLXMRglmfix(family="binomial", fixed=~Rain), k=2, nrep=5, data=rainfall)

To test the random effect model, I created the new variable ID by:

rainfall$ID <- seq(from=1, to=34, by=1)

With Flexmix, it is possible to go a step further and let the treatment variable to vary between latent groups by: 

m.mix <- stepFlexmix(cbind(Cases, Total-Cases) ~ Rain|ID, model=FLXMRglmfix(family="binomial"), k=2, nrep=5, data=rainfall)

Latent variable analysis package for R

This is another R package for latent variable analysis. I hope it can soon include multilevel and mixture modeling capacity. Another promising package is OpenMx, which is also under active development and very promising. At this moment, OpenMx seems to have more features than lavaan (e.g. categorical dependent variable, mixture model, etc.), but lavaan promises to include them in the next year or so.

Online LaTeX table generater

It probably uses some Perl or Python script to add all the formatting marks, very convenient.

Flexmix for binary dependent variable

It looks that FlexMix does not handle binary dependent variables directly, you have to transform it into "cbind(event, total-event)" format.

Convergence diagnostics with MCMCglmm

MCMCglmm does not produce more than one chain. So the way to check convergence is to run the model twice, each with different starting values, then conduct the diagnostic with:

gelman.diag(m1$Sol, m2$Sol)

Handling missing data in R

Some estimation procedures, including MCMCglmm, does not handle missing values directly. Before estimating these models, missing values need to be excluded. From this post, the command to do this is:

newdata <- na.omit(mydata)