Redistricting Court Challenge: Dr. Arrington's Voting Rights Analysis Primer

This is a good way to learn some of the terminology of the Voting Rights Act analytic techniques.  Dr. Theodore Arrington explains many of the terms.

These are rough notes from this morning.  There is a short break until 10:15.  The phone connection didn't kick in until 8:43.

Basically Mr. Walleri has been questioning Voting Rights Act Expert Dr. Arrington, who was originally hired by the Fairbanks North Star Borough to look at Dr. Handley's analysis.  He and Dr. Handley have been working together for the Department of Justice on Voting Rights cases from Texas.  The morning was focused on how voting rights experts do their analysis of the data.

The notes are as fast as I can type, which isn't fast enough.  But it will give you a sense.  There are times where I include answers in the questions and other times where I include the question in the answer - that is I don't clearly identify what part the attorney said, or what part the witness said - but you get the gist of the testimony.  There are gaps where I couldn't keep up.  I try to put question marks or three dots . . . to indicate if I'm not sure what was said or there's a gap.


January 10, 2011 Redistricting Board Court Challenge Fairbanks
Judge Michael P. McConahy, Superior Court

Dr. Arrington
8:43  begins mid sentence
Walleri:  questioning about expertise on voting rights and how many cases he’s worked on and working for DOJ - about a dozen times - and size of the VRA experts pool - about 20. 

You’re familiar with Dr. Handley?  You’ve worked with her before?
Arrington:  Yes, on the Texas case.  Next week I’ll be in DC with Texas case where DOJ is suing TExas, actually it’s Texas suing DOJ.  Four different courts involved in the Texas litigation. 
Walleri:  both you and Dr. Handley are working for DOJ on that?  Your role?
Arrington:  My role is to determine if state intended to discriminate against Hispanics.  Handley’s role is to determine if they actually did discriminate.  Intent v. Effect
Walleri:  Law degree?
Arrington:  No, political science.
Walleri:  Handley, lawyer?  But has PHD in poli sci.
Retrogression is legal term defined in the statutes.  I’m not a lawyer, so I perform analysis and let the judges determine.  Suppose you had a plan in which there were five districts where natives ha ability to elect candidate of choice. and one more where there isn’t a contest because whites vote the same. 
If you hae a new plan and also five districts can elect candidate of their choice, but doesn’t have the extra non-polarized district.  I can’t tell if that is regressive.  Court has to do that.
Walleri:  Been using effective and influence terms do they make sense to you?
We’ll be talking about a lot of stats, as I understand what you;re saying:  You can tell us voting patterns, but the legal implications is someothing different?
Arrington:  Yes
Walleri:  In terms of statistical analysis you and Dr. Handley use - therea re different types of analysis?
Arrington:  Standard ecological regression, inference, and extreme precinct analysis, Handley calls  homogeneous precinct analysis.   We can do estimates from precinct data and have to do iestimate nference from that?  “Average trunout of Natives in last decade has been X% and whites, and Natives are cohesive at Y%.  Y% for athe same candidate.  Crossover whites - whites who voted for the Native preferred candidte.  Those are all % not tied to individuals, estimates based on precinct analysis.
Walleri question:
Can I draw it on a chart?  Easier. 
    is % Native in each precinct.  So a precinct here would be a lot of natives, and on the left
vertical access you have vote for candidate.
On this chart you place a dot on each precinct here upper right  a lot of natives and a lot of votes for that candidate.   Scattered plot of this sort, it might look like this.  Putting in lot of dots.  Regression technique says, maybe there is a few outside this.  What is the straight line that is the best estimate of
Where does that regressive line cross 100% on the horizontal.  That’s the point where all the natives vote for that candidate.  On the other hand, where does it cross the zero access, that’s the best guess of the white vote for that candidate.  More complicated because you have one for each candidate.  That’s what was used in the S? case, and hundreds of section 2 cases.
Judge:  For every year?
Arrington:  For every election for every year…
Walleri:  Basic issues in regression analysis.  All Alaska Natives have common voting pattern.
Arrington:  No, the average, her figures are the average figures, but natives might differ from the average in different places.   It picks this up by having lower cohesion level.
Walleri:  Can you tell the cohesion level?
Arrington:  What it can do and what Dr. Handley has done is determine the cohesion level for each election and then the average of those.
Walleri:  Does it also …  a uniform spread of native.  Assuming that there are different types of native people or have different political preferences.  In one area there’s a concentration of Native people that have one type, this way.  Aware we have different types f antive poeple (Aware, but not studied)  XEskimo  people and Native people  Differen in concentration in this district. And the vote differently.  Would this analysis pick up this difference
Whte???
Overruled.
Walleri:  Indian people, your precinct, 100%, are Indians, and the 40 and 50% precincts are Eskimo and they vote differently.  How would this affect your analysis?
Arrington:  It would lower the cohesion, but it wouldn’t pick it up.  It would lower the cohesion numbers, but you wouldn’t pick up why.  The Census date only identifies Native.   In doing analysis experts sometimes have to go beyond that.  In Florida, Census only talks about Hispanics, but Cuban or other makes a big difference, so the experts have to go further.  In New Yourk we had to identify where the Puerta Ricans, Dominicans, and ?? are.
It can pick it up, but cohesion level is lower.  For each election using three methods, we can determine:
1.  % of Natives voting for the same candidate - high about 70%,  Jingles? case sets bar at 60%, I like 67%. 
Walleri:  Endogenous and Exogenous - can you explan these types of analysis
Arrington:  Endogenous means elections of interest to the cawe, Exoencous means other elections.
All in the last decade.
Ecological inference is much more complicated analysis.  Based on the regression, but goes much further.  Math far beyon me, just as a lawyer can use Lexus Nexus
Extreme ecological analysis - where more than 90% Native??????
Walleri:  Those types of analysis will be used over tha last decade?
Arrington:  Yes, because that’s whats relevant.  Longer is interesting, but it’s history not political science.
Endogenous most probative.  Problem is testing the new plan, there aren’t any endogenous elections to analyze because new district haven’t been used yet.  So you have to use exgoneous and apply data from old elections to new districts.  Hard to do, but that’s what we’re paid to do.
Turnout, Cohesion, Crossover are the three key factors.  Tells you whether racial block voting is happening and the level. 
???
In statewide elections, problem when you look forward to the new plan.  You’ve been involved where there are only slight changes to a district.
Arrington:  I’d hae to check, usually they get to court if they changed a lot.   If I’m trying to predict if a newv district will perform, is effective, I look to see if it’s like an old district in terms of geography.  If about the same geography, probably going to vote the same way.  But if very much changed, you can’t use the geography to see if it will perform.  Some of these districts are very different geographically, so it’s harder to predict.
Walleri:  harder to hear - Exhibits ?? and J5 the benchmark plan.  Can you tell us what you’re looking at when you say different
Arrington:  Redistricting board plan, 38, there’s nothing on the old plan which resembles that.  Starting here inf Fairbanks and going to the coast.  Part of 6, 8, old 38, very different geography.  No old district we can say would behave like the new one.  Have to use other procedures, there are other procedures, but that one can’t be used here.
Walleri:  There are other districts that are similar.
Arrington:  Find 40 for me.  This is a good example of a district you cans ee.  That’s similar geography.  On that basis alone, we can predict it will behave in the next decade as it has in the past.
Walleri:  District 36  some similarities but not all?
Arrington:  Unless similarity pretty close, have to rely on the numers.  You can look to see where the people are, not just the land. if the people are in the same place, but just the physical geography is different, then you could say, it would be the same.  Can’t just rely on the geography.
Walleri:  Doing this racial block voting, trying to predict (yes) and doing that you said, here benchmark would performt he same, but something like 39 or 38 how do you predict how these would do
Arrington:  Typically, we do what Dr H has done.  Econological regression or inference, determine average across the state, turnoutout for whites and natives, cohesion for natives, and crossover for whites, and then determine who would be the ??? to determine if effective.
Walleri:  measures of reliability of these?
Arrington:  Not really.  No statistically significant test for this.  There is a test in ecological inference but not usually applied.  Because the court itself has to make a ruling.  These techniques gives you the best estimate.  If Dr. H says 41*% is the best estimate.  That’s fine.  There may be things behind our numbers which we didn’t anticipate which causes us to make a bad call. Type 1 or Type 2 errors.  I could say it was not effective and in ten years it turns out I was wrong, or that it is effective. 
If geography the same no problem.
Walleri:  has that ever happened?
Arrington:  in my case?  no.
Walleri:  Have people ever found that with similar geography that it didn’t work out?
Arrington:  No, but after the case is over, I don’t look back.
Type 1 and 2 - predict it will be effective an isn’t, or predict isn’t but it is.
Walleri:  VRA regulations.  Part 5157-5159 Federal Reg.   page 21249,  how does statistical analysis relates to these factors, how the DOJ looks at redistricting plans.  Is the analysis you’re doing, extent to malaportioned districts right to vote for majority?? Is this what you’re looking at?  SURE  ….  Extent of fragmentation, overconcentrated, alternative plans satisfy legit government interest NO  Departs from objective redistricting criteria deviates from ???   Not what we’re talking about.
Walleri:  compactness and contiguity are not addressed in these statistical analysis?
Arrington:  yes
Walleri:  fair to say, considers some of the DOJ factors but not all?
Arrington:  yes
Walleri:  We have racial block analysis, including cohesion analysis.  What is benchmark analysis?
Arrington:  How many districts in the old plan did natives have an ability to elect the candidate of their choice, even if the choice was a native.
Wallerite:  We’ve been using the term native.  When court in Texas, they used ability district.
Arrington:  that reflects legislative language.
Wllerie:  are we saying the same thing?
Arrington:  I am.  Effective isn’t my normal language.  An ability district has to be more than a chance, but not a certainty.  There are no certainties in politics.
Walleri:  Using terms, does everyone using effectiveness mean the same thing as abilities?
Arrington?  I don’t know.  Have to ask.  In Texas court said opportunity is different from ability.  I think a legal distinction.  You have to ask. 
Walleri:  Texas court looking analysis of certain districts giving Hispanics or Blacks an opportunity to elect and DOJ was looking at ability to elect.
Arrington:  if the district might in the future offer the ability to elect because the Hispanic population is growing, Court says we have to know if they can now, not in the future.
Walleri:  Who hired you in this case?
Arrington:  FNSB - asked me to review Dr Handley’s report and testimony.  I did.
Walleri:  Did you independent analysis? 
Arrington:  Eventually I did because asked to review the demonstration plan and compare it to the proclamation and benchmark.  We have the new precincts proposed by the proclamation plan, so we can take statewide elections to see what the results were in the new districts.  Handley used that well in Texas to see how new districts would behave.  Reconstructive data analysis.
Walleri:  Looking at statewide office elections - exogenous elections - not the actual rep and senate analysis.  It can give different insights in how the districts would behave
You don’t disagree with her racial block voting analysis.  Correct.  Or measurement of cohesion?  Don’t disagree. 
What do you understand the benchmark in the old plan
Arrington: 5 effective in house and 3 in the Senate

Three answers:  most of state 41.8% but two exceptions
1.  Substantial overlap in benchmark district 6, might need more.  In HD6 in benchmark, much more racially polarized than rest of the state.  White crossover is lover.  So if big overlap between new district and HD6 benchmark of whites, then higher.
2.  In Aleutian Islands and SE where voting is not usually not racially polarize according to Handley’s analysis, we can have an effective district with lower %.

Walleri:  In these areas the Native effective, lowest, we can go to still have effective?
Arrington:  I think there is no lower limit.  I can describe a district that doesn’t have polarized voting, but don’t know how to explain that in l???.  If no polarization, then Natives will be able to elect candidate of their choice,

Walleri:  You talked about significant levels of white population into the new district.  How do you know if a portion of 6 in the new district.  How do you know which part of HD6 has the polarization?
Arrington:  Using the census, we can determine how many whites in that district that were also in old district.  We have to determine how many whites ????
Ne has 1000 whites who lived in old 6, then 48.6 is not enough.  If 100 whites from old 6, especially if a lot of natives from old 6, 48.6 would be fine.
I need to know concentration of natives plus overlap with HD6, whether have white districts, and polarized?????

Walleri:  Old HD6 has to be 49 ok half, to be effective.  We don’t exactly know in Bristol Bay what the lowest number is.  In rest of state, round to 42 or above.  In terms of looking at the new districts, which standards, we have to figure out critical HD6’s white population is going into the new district.  Does it matter if large portions of natives going into the new district.
Arrington:  irrelevant.  they have high cohesion.  It’s the whites that matter.
Walleri:  Yo agree, I think, that 6 was an effective district.  What about new plan 38?
Arrington:  By the numbers, the analysis I do, the answer is yes. 
Walllerie:  I understand from your deposition you said it was iffy.  Not ideal.  How confident just on the numbers it’s effective.
Arrington:  very confident.
Wallerei:  But using other factors?
White objection:  didn’t do any analysis of this. 

time for morning break, back at 10:05