Monday, November 22, 2010

Planning to tweak the votermedia award algorithm

As I've been watching how the contest is working at, it seems to me we can tweak one of the model's parameters to make it work a little better. This parameter affects the trade-off between two things we'd like to achieve, both of which encourage competition but in different ways:

1. Limit the maximum share each blog can win. Currently with 5 or more competitors, the maximum votable amount is 30%. This ensures that at least 4 blogs can get funding, thus enhancing competition that way.

2. Even though their shares are at or near the 30% cap, we'd still like the top blogs to feel some incentive to keep making an effort. If the top 3 are all up against a hard ceiling at 30%, they may have little or no incentive to keep working at their blogs.

A parameter that affects this is what I call the "interpolation range". We originally created it because voting is multiple-choice. If the choices are 5% apart, then a non-interpolated algorithm would have awards that jump suddenly from one choice to the next -- e.g. stuck at 20% for a while, then suddenly the next day 25%. So instead we interpret each vote as spread evenly across a 5% range, e.g. a vote for 20% is spread from 17.5% to 22.5%. This smoothes out the award response to changing support levels in a reasonable way.

It also has the nice side effect of leaving a little wiggle room for competition around the 30% ceiling. Getting more votes at 30% will gradually push a blog's award up from 28% to 29, 30, 31 and max 32, although it's very hard to reach 32%, since only one-tenth of each vote would count. AMS Confidential is at 32% now mainly because their older votes at 40% (the maximum vote when there were only 4 competitors) haven't fully decayed yet.

Looking at the shares now at, and the recent share changes that we can see at, I think we can manage this trade-off better by changing the interpolation range from 5% to 10%. Reasons:

- A 4-point range isn't much incentive for the top 3 blogs to compete. 10% range would result in about an 8% effective range of likely award shares -- more incentive to compete.
- The fourth blog would still have an easy opportunity to get some funding, but would have to work for it a little more than now -- not necessarily a bad thing.
- 10% range would take out the "air pocket" that AMS Confidential dropped through on November 18 (see -- the drop would start earlier and end later, becoming more gradual, which seems to me more realistic, and less an artifact of the algorithm.

I tested the calculation yesterday. It would have changed yesterday's awards as follows:
- AMS Confidential up 2% from 32% to 34%
- 11' Eleven" up 1% from 30% to 31%
- Tyler's Blog down 3% from 9% to 6%

I wouldn't focus too much on these specific numbers. At other times in future it's likely to play out differently. Again, AMS Confidential's lead is mostly from old 40% votes that are decaying away anyway. I've been happy to see a fourth challenger (Tyler) moving up, and I'm not eager to slow him down. But I think as a general long-term principle of competition, this change would be healthy.

I phoned Tyler and explained all this to him. He said he doesn't have a problem with it. (Thanks Tyler!) I've also briefed the AMS Elections Administrator, Erik MacKinnon (email elections[at]

Your comments welcomed here or by email to mark[at] I plan to phase this in tonight unless I hear some compelling objections. Since this is an experimental contest, we've reserved the right to make changes as we go along -- see Terms.

No comments: