Page Strength

“There is a great satisfaction in building good tools for other people to use”. Freeman Dyson (Professor Emeritus, The Institute for Advanced Study) in “Disturbing the Universe” (1979) Chapter 1.

I think Matt Inman and Rand Fishkin would agree with Professor Dyson, which is why they are probably very pleased with their new tool, Page Strength. The real question though is “Is it a good tool?”

Rand says “The Page Strength Tool is designed to replace the often inaccurate and infrequently updated Google PageRank score in the toolbar”. However the first thing I noticed when I plugged in a couple of urls was how little the Page Strength differed from the PR and mostly by similar amounts. So I collected a few random urls during the course of a day and plotted the Page Strength against the PR on an x,y chart. The results were as follows:

Page Strength vs PageRank on an x,y chart.

On this set of data which is 26 urls the correlation coefficient is 0.92 which is extremely high. This is not so surprising in that they are both dimensionless variables heavily dependent on link counts. It would be interesting to know what the correlation coefficient is with a larger sample set. This is something Rand or Matt should be able to calculate easily since both values for any given url are kept in their database.

PageRank is a lousy predictor of SERPs position because it is keyword independent. However when I saw that Page Strength included at least one keyword related factor in its calculation (position in Google for the first four words in the title tag) I though it would be interesting to see how much better Page Strength was than PR as a predictor. There are lots of ways at looking at this but as all I wanted was a rough indication I looked at the Page Strength and PR values for the first 25 urls in the Google search for |engagement rings| and compared their correlation coefficients. The results looked like this:

Page Strength and PageRank as predictors of SERPs position.

PageRank as expected does indeed have a poor correlation with position in the SERPs with a correlation coefficient of 0.441 but Page Strength is even worse with a correlation coefficient of 0.123. On different searches the results would vary and of course the relationship may not be linear as assumed here but despite the statistical liberties Page Strength does not look anything like a possible contender to replace PageRank as a metric.

However there is one very good lesson that can be learned from this not so good tool. Barely two weeks since its launch Rand’s website has obtained hundreds of inbound links from sycophantic bloggers. A very good example of Link Bait in action and a lesson for us all!


  1. bobmutch said,

    September 12, 2006 @ 8:10 pm

    Very interesting analyzation!!! I think perhaps the tool would have been better if they had left the 0 to 10 scale behind. I would be very interested to see how my ILQ rates compaired to PR.

    The ILQ is from 0 to 80 million.

  2. duz said,

    September 13, 2006 @ 6:09 am

    Bob I took a look ILQ compared to PR for you, using the data from your site General Web Directories Rated by Inbound Link Quality (ILQ). The first chart using all the data where PR ≠ 0 shows a weak correlation.

    PR and ILQ all data

    I am sure that this is due to the extreme values for a few of the results. If these are removed from the data set there is no correlation.

    PR and ILQ partial data

  3. bobmutch said,

    September 13, 2006 @ 5:57 pm


    With the list of directories there is the top 6 that range from 5,800,000 to 44,000 and then the rest are from there it is 10,000 to 42. This shows us that there are a few very high ILQ directories and then a bunch of low ILQ directories.

    I would be interest to see a graph on the directories from 42 to 10,000 and see what the correlation coefficient is. Looks like your list is from 55k to 58k.

    Also I am wondering if Rand’s good use a 0 to 100 million scale whether the correlation coefficient would change.

  4. duz said,

    September 14, 2006 @ 6:00 am

    Bob - The second chart (r=0.8) shows what you want, the correlation for the data points for ILQ values 0 to 6,000.

    >Also I am wondering if Rand’s good use a 0 to 100 million scale whether the correlation coefficient would change.

    If the method of calculation remained the same then it may be different but not by much because the distribution would be similar.

RSS feed for comments on this post · TrackBack URI

Leave a Comment