[EM] Re: Automated districting

[Ernest Prabhakar]

Hi Matt,

On Jan 10, 2004, at 3:50 PM, matt at tidalwave.net wrote:
> However, the only objective consideration in drawing census tracks 
> that I am aware of (in the U.S.) is that they contain 5-10,000 
> persons, which leaves census tract drawing vulnerable to political 
> manipulation.

I think this number is actually the crux of the issue.   If the size of 
a census tract is far smaller than the size of the district, then I 
would argue it doesn't matter how much bias there is in how the tracts 
are defined.  I think we can count on at least 10 tracts per district, 
as a conservative assumption.

Think about what it means to gerrymander.  It means slicing up a 
district so that there's a solid majority for your party.   At the 
state level, it means to slice up the districts so you have a solid 
majority in as many districts as possible, even though that means 
giving the opposing party a solid majority in their districts (hence 
the polarized politics of California).  Ideally, you want as small a 
lead in each district you 'control', to allow you to control more 
districts.  This means precise control over 'edge' regions, where 
there's an overlap of (say) Republican and Democrat populations.

To gerrymander with tracts, that means you'd need to either:
- a) know the results of the algorithm in advance, so you can bias it
- b) try to bias each individual tract

The whole point of an automated algorithm is that it is setup *before* 
the census.    The reason we have census is that we don't actually know 
what the population is, so there's a great deal of variation.   Plus, 
the algorithm should be designed with a high degree of sensitivity to 
small fluctuations, so that it is infeasible in practice to predict in 
advance whether a given 'edge' tract will be part of District A or 
District B.  A gerrymandered would have to guess whether to bias a 
given tract for or against his party, and could easily end up guessing 
wrong.

If one really wants to eliminate all opportunity for bias, then I think 
we should just go with Mike's prescription for rectangular districts.  
However, that algorithm -- because it is so predictable -- will let 
parties decide ahead of time whether they'd win or lose under the 
scheme, so it could lead to a nastier adoption battle.

> Also, bandwith alone as an alternative ignores the difference between 
> expressways and small roads.  Some roads may have many exits inside a 
> census tract others may have none.  Using Miles per hour as a 
> compromise measure fails to distinguish between roads with many 
> traffic lights or stop signs and those with none.  I don't think roads 
> can be utilized as an unbiased or accurate measure of community  
> relatedeness.

I actually think it is a good thing that expressways and small roads 
count the same.   Also, its a fairly easy rule to say "count an 
expressway if it has an exit in this tract, otherwise not."

I'm not say such a rule is totally unbiased - nothing really is, if you 
think about it, since every model makes some implicit assumption about 
what's important.   I'm just saying that the biases caused by the 
nature of roads will NOT positively correlate with the goals of 
gerrymandering - its just too difficult to predict. The only way I 
think you could use this in your favor is if you could show there were 
gross correlations - i.e., people living near freeways were more likely 
to be Democrat.

This question could probably be settled by empirical calculation.   
We'd need a fine-grained grid describing Republican vs. Democrat, then 
impose a coarse-grain grid for each of the tracts.   The challenge 
would be to see if it was in fact possible to redraw tracts, or build 
additional roadways, to create a gerrymandered district.

If someone was willing to precisely define the conditions, I could 
probably code something up.

-- Ernie P.