In a recent project for a class, I statistically analyzed various political and science blogs. One of the experiments was determining whether a simple Bigram model could accurately discriminate between pro-Darwin and pro-Design blogs. What I found was that it is possible to achieve high classification accuracies, as long as one constructs the model in the right way.

One approach is to classify a blog based on its nearest neighboring blogs. A byproduct of this approach is the ability to rank blogs based on a test blog. Listed below are 20 different blogs (10 pro-Darwin & 10 pro-Design), ordered according to log-probabilities which indicate the "closeness" of each blog to the test blog. In this case, the test blog is iDesign. It's interesting to know which blogs are most similar to our own blog.

idesign on model idesign: -2620.211072369357
idesign on model id-in-the-uk: -7268.4689091410055
idesign on model evolution-blog: -7375.275833535828
idesign on model creation-evolution-design: -7426.824936246623
idesign on model evolution-list: -7453.481304947268
idesign on model evolving-thoughts: -7689.827741324608
idesign on model id-plus: -7739.829580298751
idesign on model red-state-rabble: -7761.04457329062
idesign on model id-update: -7912.984027203059
idesign on model uncommon-descent: -8048.212928914839
idesign on model design-paradigm: -8213.025938207627
idesign on model stranger-fruit: -8253.903617022663
idesign on model telic-thoughts: -8322.958305060785
idesign on model pandas-thumb: -8565.82807971049
idesign on model id-the-future: -8591.20770360031
idesign on model loom: -8823.779572725092
idesign on model evolution-news: -8930.788791866384
idesign on model ncse: -8964.343376093331
idesign on model pharyngula: -9061.888668178386
idesign on model dispatches: -9318.132985935674

Interestingly enough, 4 out of 7 of iDesign's nearest neighbors are pro-Darwin blogs! I guess the main conclusion to take away from this data is that iDesign is a well-balanced blog that seeks to present both sides objectively.