Monkey Cage: Donald Trump did not win 34% of Latino vote in Texas.

[jimrtex]

Thoughts?

Quite sloppy analysis.

The data set is truncated, and it appears they used the truncated set for their analysis. Alternatively, they used the entire data set, and then when they realized they lost some of their data, fudged the article.

Their classification of large, medium, and small counties is totally moronic. Hidalgo is not a small county. Cameron is not a small county. Webb is not a small county.

Why would anyone classify El Paso as a large county, Collin as medium size, and Hidalgo as small. That is something a junior high student would do, not someone who is supposedly a doctoral student.

Rather than attempting to make a representative sample, they selected certain counties. Even if they include 75% of Hispanic VAP, they do not necessarily include 75% of Hispanic CVAP, or 75% of Hispanic voters.

Higher income, or more integrated Hispanic voters would be more likely to vote for Trump, and may have been carelessly excluded from the study.

[jimrtex]

There's no way that Trump won 34% of the Latino vote in Texas and I'm not sure why Republicans are keen on arguing against this. Shouldn't you guys be happy that you can win national elections without feigning concern for immigrants and their spawn?

Did you read the article in the Washington Post?

Did you download the data set that they provided?

Why aren't you guys be out ... (whatever)?

You must be logged in to read this quote.

"In rural counties, Clinton won an estimated 77 percent of the Latino vote against 19 percent for Trump."

They treated Webb, Presidio, Starr, Hidalgo, Zapata, Potter as rural counties.

Look at the swing in South Texas.

This study purported to be by a doctoral candidate at UCLA. It was junior high level incompetent. If you are in junior high, no personal insult intended.

[jimrtex]

No way. The Rio Grande Valley would be freaking red and Houston would've voted for Trump heavily if it were that case. Also the state overall swung more Democratic.

I'd say it was near 25-30%. I'd say the 9 point gap victory for him in the state is more fueled by more White Texans voting more for Trump, as he is very popular among White Texans.

The state overall did swing Democratic. Like everywhere else, the swing was in the more populous counties.

"Instead, we draw on 4,372 precincts across Texas. These precincts cover all regions of the state and more than 75 percent of its Latino population. You can download our data here."

If you download the data set, you will indeed find that it has 4372 precincts. It also stops in Travis County, though they claim to have included Webb and Zapata. So either they did not include Webb and Zapata in the study, or they did, and then lost part of the data, and saw that they only had 4372 rows, and fudged their report.

"Some analysts looking at county-level data think that rural Hispanic voters shifted toward Trump, as Geraldo Cadava argued,"



Note the counties listed and the percentages in the next paragraph.

"In rural counties, Clinton won an estimated 77 percent of the Latino vote against 19 percent for Trump. In medium-size counties, such as Corpus Christi, Lubbock, Amarillo and Plano, Clinton won an estimated 73 percent of the Latino vote — still 12 points higher than the exit poll estimate of 61 percent. And in large cities, such as Houston, San Antonio, El Paso, Dallas-Fort Worth and Austin, Clinton won an estimated 80 percent of the Latino vote, while Trump won just 14 percent."

Webb, Presidio, Starr, Hidalgo, Cameron, Zapata, and Potter are not rural counties.

Hidalgo, Collin, and El Paso counties are quite similar in population, yet the study classified one as large, one medium, and one small. Corpus Christi, Amarillo, and Plano are not counties. And besides they classified Potter as a small county, not a medium county. But they did classify Randall as medium. Maybe I should give them half credit: half moronic and half idiotic.

The counties they did include (at least in their data set) are Bexar, Cameron, Collin, Denton, El Paso, Fort Bend, Harris, Hidalgo, Lubbock, Nueces, Potter, Presidio, Randall, Starr, Tarrant, and Travis.

They did not include Denton, Ellis, Montgomery, Brazoria, Galveston, Williamson, Comal, Guadeloupe, Atascosa, Zavalla, Dimmit, Tom Green, Ector, Midland, etc.

Imagine that you were doing a study of the Pennsylvania elections and you wanted 33% of the population. So you choose Philadelphia, Bucks, Montgomery, Delaware, and Chester. You find that 90% of "Pennsylvanians prefer the Phillies, Flyers, Eagles, Villanova and LaSalle to the Pirates, Penguins, Steelers, and Penn State."

See any problem with the sample?

They picked a sample that included 75% of Hispanics, but which is not representative of 100% of Hispanics.

Their charts show "net Clinton" - "net Obama", which is presumably: (Clinton - Trump) - (Obama - Romney), but that is not how swing is measured. If Obama had 75% of the vote, and Clinton got 70% there might be a net increase in the margin, but not in the share of the vote.

"In the 864 precincts in which 75 percent or more voters are Hispanic, Clinton won more votes than Obama in 723 of them, fewer votes in 130, and tied in 11.

If we compare Clinton’s vote margin over Trump to Obama’s margin over Mitt Romney four years ago, Clinton had a higher margin than Obama in 692 of these 864 precincts — or 80 percent. "

Using their data, Clinton's margin of victory was greater than Obama's in 696 of 867 precincts (with greater than 75% Hispanic population). This data set at least got to Webb County.

However, Clinton's share of the two-way vote was only greater in 548 of those 867 precincts. Trump improved on Romney's performance in 37% of the heavily Hispanic precincts.

Trump only improved on Romney in 18% of the precincts that were less than 25% Hispanic.

Did they take into account black voters?  Did they take into account citizenship? Non-citizens likely favored Clinton, but how many voted?

The study is total amateur hour. Maybe UCLA should be merged into LACC.

[jimrtex]

Yes, of course, but I wouldn't underestimate the difficult of gathering such a large data set; this is a very labor-intensive, time-intensive process and I know because I've conducted studies myself. It took me a very long time to gather the data. I'm mostly happy that someone did this.

"Francisco I. Pedraza is assistant professor of political science and public policy at the University of California at Riverside. Bryan Wilcox-Archuleta is a political science PhD student at the University of California at Los Angeles. Additional graphs are available."

Pedraza has a PhD from University of Washington and Wilcox-Archuleta is a doctoral candidate at UCLA. UW and UCLA are reputable schools. This is in their field of supposed expertise. Their article is crap.

Are you familiar with the expression "soft bigotry of low expectations"?

[jimrtex]

I agree with jimrtex observations. A study that puts Hidalgo in the small category looks suspect. Hidalgo is the seventh largest pop in TX and is larger than El Paso, which is in the large category. Also the study relies on ecological inference, which is not a simple methodology and one needs to take care to account for changing demographics over time among other things to get it right. If the study doesn't understand the size of the counties I would subject it to added scrutiny.

One the basic point, it is quite possible to get move votes in a category and still have the percentage of votes in that category decrease. The cause would likely be due to the overall increased turnout which changes the denominator as well as the numerator.

I downloaded their data set that compares 2012 to 2016 results.

I can almost replicate their claim that 692 of 864 counties that were 75% or more Hispanic had a greater net Clinton vote than net Obama (I get 696 of 867). But this uses

(Clinton-Trump) > (Obama-Romney).

But if you instead use:

Clinton/(Clinton+Trump) > Obama/(Obama+Romney)

Then the number drops to 548. In 38% of heavily Hispanic precincts, Trump outperformed Clinton.

Look at the swing map for Texas and notice Zapata, Starr, Duval, Zavala, Dimmit, Atascosa.

For the 2012-2016 comparison they screened precincts that they couldn't match. In some counties, 90% of precincts matched, in others none. But there is no guarantee that a precinct with the same number is the same precinct. Harris County recycles precinct numbers. And in other cases if a precinct number in 2016 was not present in 2012, it may be due to a precinct split. Thus there is likely a demographic bias in their selection of precincts that are comparable.

[jimrtex]

Yes, of course, but I wouldn't underestimate the difficult of gathering such a large data set; this is a very labor-intensive, time-intensive process and I know because I've conducted studies myself. It took me a very long time to gather the data. I'm mostly happy that someone did this.

"Francisco I. Pedraza is assistant professor of political science and public policy at the University of California at Riverside. Bryan Wilcox-Archuleta is a political science PhD student at the University of California at Los Angeles. Additional graphs are available."

Pedraza has a PhD from University of Washington and Wilcox-Archuleta is a doctoral candidate at UCLA. UW and UCLA are reputable schools. This is in their field of supposed expertise. Their article is crap.

Are you familiar with the expression "soft bigotry of low expectations"?

Uh, this isn't an academic publication, it is a study for an organization that's probably a side-project for these people.

Unless you have a critique of the estimation itself, what you're doing is nitpicky and childish. I fail to see how their categorization of counties matters. It's totally irrelevant to the goal of this study, which is to estimate how Latinos in Texas voted in 2016.

Their claim to expertise is that they are PhDs in Political Science, with a specialization in racial and ethnic studies. Their article is something that they might well seek to craft into an academic publication. Wilcox-Archuleta might use it for his dissertation.

They claimed to have used a sample that had 4372 precincts, and included Webb and Zapata counties. They did not include Webb and Zapata counties in their data set, but the data set had 4372 rows. A study of that type that excludes Webb County is not credible.

The Pedraza article you are attempting to defend linked to this article.

https://www.washingtonpost.com/news/monkey-cage/wp/2016/11/17/rural-hispanic-voters-like-white-rural-voters-shifted-toward-trump-heres-why/?utm_term=.08cff2a8bc2d

That article by Geraldo Cadava challenged the poll by Latino Decisions because it concentrated in urban areas.

The purpose of the Pedraza article was to disprove the claim of the Cadava article. To do so they claimed to segregate their Texas data into large, medium, and small counties. This is their classification.

Harris           3999565  Large
Dallas           2406051  Large
Tarrant          1807048  Large
Bexar            1711320  Large
Travis           1024257  Large
El Paso           800647  Large
Collin            778443  Medium
Hidalgo           761203  Small
Fort Bend         585205  Medium
Cameron           406220  Small
Nueces            337178  Medium
Lubbock           278818  Medium
Webb              250304  Small
Randall           148278  Medium
Potter            121073  Small
Starr              60968  Small
Zapata             14018  Small
Presidio            7691  Small

I agree that Presidio and Zapata counties are small. Pedraza/Wilcox-Archuleta do not appear to have included Webb and Zapata in their study, So their "small" category consists of Hidalgo, Cameron, Potter, and Starr counties. 86% of the population in their small county category is from Hidalgo and Cameron county.

Now look at Potter and Randall counties. Amarillo spans the county line. The Pedraza article classifies Amarillo as a medium "county". They similarly call Corpus Christi and Plano "counties". They got lucky with Lubbock since the city of Lubbock is in fact in Lubbock County. Amarillo began in Potter County, and has generally grown southward towards I-40. US 66 wound its way into Amarillo, while I-40 went more or less east-west. I-27 also goes southward from Amarillo.

In 1950, Potter had almost 6 times the population of Randall. By 2010 they were almost equal. Randall might have more than Potter by now (2016).

But Pedraza shows Randall with 28,000 more persons. So I looked at the Randall County data. They duplicated six precinct (111, 208, 305, 324, 325, and 381). Precinct 208 is included 3 times!.

I apologized to junior high students, for characterizing the Pedraza article as being at their level.

[jimrtex]

The argument that the total number of votes Clinton received in heavy Latino districts exceed that of Obama 2012 means that the share of the Latino vote for Clinton in TX must be higher misses several key facts.

1) Total eligible voters in TX increased 8.8% from 2012 to 2016
2) Turnout in 2016 is higher (51.6%) than 2012 (49.2)

It is also possible and in fact likely that the share of the Latino vote in TX was higher in 2016 than in 2012.  So it is totally possible for the Clinton share of the Latino vote in TX decreased from Obama 2012 while the total number of Latino votes increased from 2012 Obama.

That's obviously a possibility but the data doesn't really point to this!

The Trump share of the vote increased in 38% of the counties that were 75% or Latino.

If we are trying to determine whether Trump's share of the Latino vote increased over that Romney, we certainly do want to look at swing:

    Trump/(Trump+Clinton) - Romney/(Romney+Obama)

And not change in net difference:

    (Trump-Clinton) - (Romney-Obama)

[jimrtex]

Latino decisions been at it for a while now.  See

http://www.latinodecisions.com/blog/2010/11/15/proving-the-exit-polls-wrong-harry-reid-did-win-over-90-of-the-latino-vote/

Where they argued that exit polls showing Sharron Angle in 2010 NV Senate race won around 30% of the Latino vote are wrong that that her share of the Latino vote was in the single digits.

Not saying they are wrong but clearly this is an organization with an agenda of increasing the relative power of Latino interests within the Dem party arguing that the Latino share of the Dem vote is higher than what exit polls imply and with it the relative clout of Latino pressure groups in Dem power structure should increase as well.

If they are right then of course then GOP share of white vote is even larger than the exit poll suggest.

Doesn't the last chart predict that if a precinct were over 60% Hispanic that Reid would receive over 100% of the vote?

Reid did get close to 100% in precincts with almost no Hispanics. So they had to be in North Las Vegas.
And it appears that we could get a much better fit with a second degree fit.

[jimrtex]

You must be logged in to read this quote.

Speaking as a Travis County resident, this is not a difficult question. If your survey of Hispanics in Texas is limited to Hildago, it is by definition unrepresentative of Texas as a whole. If your question is, "why are Latinos in Texas going Trump", then it behooves you to go into places like Travis.

Hildago tells you very little about the swing. I don't go to Wyoming to explain the 2008 Obama election. Same principle here.

Now, if  I were doing a poll with serious systematic issues to reassure people that yes, Hispanics are still democrats, I would do exactly what they have done here.

Dead Flag Blues had misunderstood something I said about Travis County.

The data sets are sorted by county and precinct. They have an ID comprised of the County ID (a three digit number, and a four digit precinct ID). When I was looking at the data set, I was converting the numeric IDs to county names. That was when I discovered that Webb and Zapata were not present, and Travis happened to be last alphabetically.

I had conjectured that someone had messed up a cut and paste (e.g. instead of selecting all the rows had somehow selected only to a certain row, and then cut and pasted). This would have made Travis County a partial data set. Since Travis County numbers its precincts by commissioner district, this could have introduced a geographic and demographic bias (e.g. CCD 1 Black; CCD 2 Anglo Liberal; CCD 3 Republican opportunity; CCD 4 Hispanic).

But there was a second data set, that combined 2012 and 2016 data, and it went through Webb County (still no Zapata). It showed that all of Travis County had been included.

The total populations of the counties that I listed in an earlier post was from their data set. They are similar to the census population but not exactly. After I saw that they had included precincts multiple times for Randall County, I decided not to try to figure out where they had messed up.

But their data excluded counties such as Williamson, Comal, Guadeloupe, Medina, Atascosa, etc.

[jimrtex]

The issues regarding the Rio Grande Valley remind me of the issues I ran into when making my 2012 White Obama Lovers by County map last year.

With Texas specifically, I had to continue tweaking, and tweaking, and tweaking...and tweaking my formulas to get something that looked believable for most of rural Texas. In most states, a rather uniform formula could be used; where there are Latinos, it doesn't work. With TX in particular, it was even worse. I had to create a whole range of scenarios to balance out everything.

In the northern part of the state, I kept running into issues with white support for Obama being negative - even when bumping down Latino turnout and support for Obama to ungodly levels even by Latino standards. There are still counties on the map where white Obama support is at like 1-2% after all of that, including bumping Latino support for Romney up to 60-70% in some cases.

In South Texas, it was the opposite. White support for Obama kept producing ungodly levels of support that were showing up as majorities in some counties. It doesn't help that white population in some of these counties is so small that moving Latino support by 5-10 points shifted Obama's white support by 50 points. I continued to increase Latino turnout and (mostly) support to levels comparable to black voters in order to get something halfway realistic. Even then, the share of white support on the map still sticks out compared to the rest of the state, lingering between 25-30%.

(It's worth noting that this broader trend - whites seeming to be more in favor of Obama in the more heavily minority rural counties than in neighboring rural counties with smaller minority populations - was something I ran into to varying degrees quite consistently throughout the country, including in heavily Black and Latino areas, the former of which is much easier to project in terms of turnout/support)

Anyway...I'm wondering if there might have been some truth/accuracy to my initial projections for the map, in that whites were actually unusually Democratic in these heavily Latino counties and that - when combined with drop-offs (potentially? haven't looked yet) in Latino turnout in these areas - a huge swing among the otherwise small white vote to Trump there was enough to swing the Rio Grande overall to Trump, even though Latinos may have been more Democratic this time around and/or turned out in larger numbers? It's worth noting again though that in many of these Rio Grande counties, I had Obama at 90% among Latinos...so I'm not sure how they could have gotten much more Democratic there than they were in 2012.

And maybe that's why there's a swing to Trump there: they didn't, but the whites swung to Trump by a very large amount and Latino turnout (at least in raw numbers) was either down or there wasn't enough of an increase to cancel out white swing. I do have a hard time believing Trump improved among Latinos overall, whether in TX or nationally.

I don't know if you are interested, but there is precinct-level data for Texas, including demographics on the Legislative Council redistricting web site.

I assume you attempted to disentangle race from ethnicity. Most Texas Hispanics identify as white.

The 2006 special election for TX-23 may be indicative of what you are saying. Because the new map was imposed in 2006 after the primary, elections were run as a special election concurrent with the November general election. Special elections in Texas are run without primaries, with candidates from all parties running. If there is no majority, a runoff is held. This happened in TX-23.

Henry Bonilla received 48.61% of the vote against 6 Democrats and an independent. In the runoff a month later he was defeated. Generally, he received a larger share of the vote in the runoff than in the special election. Republicans turnout is typically higher in runoff and special elections, and in presidential rather than gubernatorial elections, which of course is repeated elsewhere.

Overall turnout was about 60% of that for the gubernatorial election. But in some counties, more votes were cast in runoff than the gubernatorial/special election.

For example in

Zavala County was

438 Bonilla 871 all others, total 1309 in the special/general
255 Bonilla 1280 Rodriguez, total 1533 in runoff

The falloff for Bonilla was typical between the special/general and the runoff. But a 40% increase is huge on top of an expected dropoff.

Rodriguez was re-elected in 2008, but defeated in 2010 on lower turnout.

Gallego was elected in 2012 in the new district, but was beaten in 2014 on lower turnout. What was quite remarkable was that Hurd was re-elected in 2016. Even more remarkable was that Trump ran ahead of Hurd in many counties (this in part may be due to voters picking the Spanish-surnamed congressional candidate, and Clinton not being relatable to on a personal level.

Factors such as the TX-23 race will provide local variation in turnout. In South Texas, the Eagle Ford may also cause voters generally to reject a Democratic Party that is hostile to energy production.