Inventing a Metric

By on in , with One Comment

As I’ve written about on many occasions, I’m a huge proponent of establishing clear objectives before trying to come up with metrics. I’ve always claimed that, with this approach, you really get freed up to come up with truly relevant metrics, rather than settling for metrics that are what you’ve always measured or metrics that are easy to get at.

An interesting example came up on the webanalytics Yahoo! group yesterday. The original post:

I use web analytics mainly in the context of natural search. One thing that I always want to measure is `landing page diversity’ and would like to know if anyone can help me with a metric, or metrics, tomeasure it.

To explain a little more fully: if a site has 10 pages, then at one extreme all visitors enter through the home page (zero diversity) andat the other extreme 10% of visitors enter through each page (100%diversity).

But how do I put a number to all the variations in between, in effect differentiating between sites in which a few landing pages attract most of the traffic and ones where a higher proportion of internal pages attract traffic.

Any help greatly appreciated.

While this post doesn’t explicitly state objectives, there is clearly some real thought and rationale behind what the poster is trying to look at.

A very frequent (and high quality) member of the group, “Wandering” Dave Rhee, responded, and his response triggered a thought that I posted (Dave coined an acronym — SMP, for “Smart Math Person,” which I then referenced in my response). Frankly, I was pretty tickled with the idea. My response:

I am NOT one of the SMPs on this list. But, WDave’s second, simpler thought somehow knocked down a small, distant, cobweb-encrusted door inthe remote reaches of my brain.

It got me to thinking of the Herfindahl-Hirschman Index (HHI), which many economists use as a way to measure competition in an industry (aswell as to quantitatively determine when a monopoly exists). The formula is pretty simple: you take the market share of each of the companies in the market, square it, and then add them all together. Actually, you usually sort in decending order of market share and then square and sum the top n companies’ market shares. An HHI of 1 is a perfect monopoly. An HHI that approaches zero has extreme diversity/competition.

It seems like this approach might work as a simple, yet valid, landing page diversity metric. Each landing page has a % of the “market” of entry to the site. Square that percent for each page and then add them together (within reason – the impact on the HHI steadily decreases…maybe even exponentially or logarithmically, but I’m no SMP- the farther you go down your list).

More details on HHI at: http://en.wikipedia.org/wiki/Herfindahl_index

There has not been any further discussion of the topic since my post, which can be interpreted any number of ways. My reason for reposting it here is really to illustrate how open you can and should be to establishing metrics once you know what it is you care about. The webanalytics group is a great forum that many members use for just this purpose. For instance, there’s a very active thread going on right now debating how best to measure “proactivity” in social media.

I’m a fan of the approach!

One Comment


  1. Pingback Gilligan on Data by Tim Wilson » Inventing a Metric — Redux

Leave your Comment


« »