Reflecting on Berkeley Data Science

During the past 18 months, I served as an academic officer ("curriculum coordinator") in what was then known as the UC Berkeley Division of Data Sciences, and which will be one of the elements of the new Division of Data Science and Information. I left the Division in early December to switch to a research role across the bay at UCSF.

My involvement with what became the Division stretches back to Fall 2015, when Cathryn Carson brought me on to teach one of the 6 inaugural Data Science Connector courses that term. Connector courses are usually small seminar-style, half-time offerings (2 units rather than 4) that aimed to connect concepts in the foundational course at Berkeley, Data 8, to topics in academic fields. I taught a connector entitled "Health, Human Behavior, and Data" during that first term of Data Science at Berkeley, and again in Spring 2016. During the academic year 2016-2017, I taught statistics and econometrics at Mills College in Oakland.

Looking back on my involvement with Data Science these past several years and over the arc of my own education and career, I feel a sense of great pride and awe at what the group at Berkeley has accomplished. As a sophomore in Fall 1993, I studied computer science in the introductory course COS 126 taught by Robert Sedgewick at Princeton 25 years ago, during the same term I studied econometrics led that fall term by Henry Farber. I had my hands full. There were many Sundays when I didn't get to sleep until very late, working on problem sets for both classes, sometimes via a dialup modem on a Mac SE from my dorm room.

In 1993, econometrics was its own thing, and computer science seemed like it was all about loops and "pointers" whatever in the world those were. But Stata lived on three UNIX mainframes that all of us had access to, and there was some synergy to be had in studying C++ and applied econometrics in Stata at the same time. I definitely learned plenty of UNIX along the way.

I wish that modern Data Science had been around back then, and it thrills me to see it here now, at last. No joke, I earned a C in COS 126 that term, although I got an A in econometrics. I suppose a C+ in COS 126 would have been somewhat more appropriate! Back then, it was tough to straddle both worlds without destroying one's GPA. Now, it is far more seamless.

Students in Berkeley Data Science certainly include many folks who are Comp Sci whiz kids. But they also include many who were a lot more like I was in 1993, along with folks even further toward the social side of social science, and people in the humanities. Berkeley has figured out a great way of planting poles so widely that the tent is truly big. It has been a great pleasure watching and helping it unfold.

All (immigrant) kids are always expensive

Another inquiry came in from the media that went something like this:

"I'd like to know the impact on state and local finances of an arrival of a skilled immigrant, as opposed to an immigrant with less than a high school degree."

There is some variation in the costliness of kids to governments according to the parents' education and language skills, but a lot less than there is in the net costliness of adults (i.e., how much they pay in taxes vs. absorb in benefits).  Kids are kids, pretty much.  States and localities have to hire K-12 teachers to take care of them regardless of whether their parents have Ph.D.'s or haven't graduated high school.

A wall doesn't remove the unauthorized

There seems to be continuing interest, likely because of the looming fight in Congress over the wall funding, about a FAIR report pegging the annual budgetary net costs of unauthorized immigrants at $100 billion. Folks who are in favor of building a wall argue that because a wall costs less than $100 billion, the wall will "pay for itself."

Fact checkers have emailed and called me about this, and my heart goes out to them, because the problem here really isn’t the imprecision of the facts per se.  (They are indeed imprecise.) 

The real problem is that comparing one-time costs of constructing a border wall to annual costs of net government benefits paid in excess of taxes received is a really silly comparison.  That’s because building a wall doesn’t magically remove unauthorized immigrants already in the country.  Only deportations do.

Without expanding the funding of removal, all you get is an expensive wall and 11 million unauthorized immigrants still living in the country and still absorbing more in benefits than they pay in taxes, probably to the tune of about $50 billion per year, mostly paid by state and local governments in the form of K-12 education.

But don't get me wrong, I'm not in favor of displacing 11 million people. The human and economic costs would be very large.

It's just that it's silly to compare the costs of a wall with the costs of something that a wall would not reduce.

Passel and Cohn on unauthorized immigrants

Another impressive installment from Passel and Cohn at Pew on unauthorized immigration. So much to digest, but a striking thing was how unauthorized men at working ages declined in proportion and numbers, while numbers didn’t decline for women https://t.co/c6GrDZXoC7

— Ryan Edwards (@RyanDEdwards) November 28, 2018

Van Hook on Census questions about citizenship

In case you missed it, Jennifer Van Hook looked at CPS data from 2014-2018 (through March 2018) and testified in Manhattan yesterday about the changing patterns of response that appear to be associated with its citizenship questions and the political environment.

Here is my tweet about her killer chart:

Killer chart inbound: Nonresponse to CPS among non-citizens was already elevated, and in the Trump era this disparity has widened pic.twitter.com/KeoXNi74k3

— Ryan Edwards (@RyanDEdwards) November 6, 2018

Fiscal costs of unauthorized immigrants

Several media outlets have recently expressed interest in the fiscal costs to governments of unauthorized immigrants. Although easy to conceptualize — the costs of government services and benefits provided minus the taxes and user fees paid — it isn't easy to figure out what the bottom line is for a population that is hard to measure.

A 2013 Heritage study by Jason Richwine and Robert Rector currently provides the most precise look at this. It's based on the 1997 report by the National Academies and some very reasonable extrapolation. Using the 1997 NRC methodology, they estimate that 3,444,955 unauthorized immigrant households in 2010 (Table 2) were responsible for an annual deficit of $14,387 per household (page vi, first column).  In total, that amounts to about $50 billion.

Another estimate can be derived from the 2017 report by the National Academies (bottom of page 406, Table 8-1 on page 389). For all first-generation immigrants, the statistic is something like a $279 billion annual deficit for the 55.5 million first generation immigrants and dependents. That includes all kinds of old-age benefits that unauthorized folks wouldn’t be able to absorb, so it’s too large a cost number but still illustrative. (Another bias works toward attenuation because unauthorized immigrants are somewhat less educated than all immigrants.) But if we were to prorate it, then the cost associated with 11 million unauthorized immigrants  would be 11/55 or about 20% of that, or roughly $55 billion.

So my vote is for about $50 billion. That's a rough estimate of the annual fiscal cost associated with unauthorized immigrants' absorption of benefits in excess of their tax payments at all levels of government. Most of the impact is at the state and local level in the form of K-12 education spending.

In comparison, there is a statistic of $100 billion floating around the Internet that probably comes from this report by the Federation for American Immigration Reform (FAIR).

Here are some takeaways:

• It is likely that $100 is too high

• Even if it were accurate per se, an annual "cost of unauthorized immigration" is a deceptive statistic because it is not the net cost of a policy action
  There is no way to recoup the $50 billion absorbed by unauthorized immigrants annually with a costless policy. A more useful statistic to guide policy would be the net cost of removing 11 million unauthorized immigrants. One estimate places the cost of removal alone at about $100 billion, leaving aside the added enforcement costs of maintaining removal once it has occurred. Combining these produces a net additional cost to taxpayers of $50 of removing all unauthorized immigrants in a year (if that were possible): the $50 billion gained by not paying benefits in excess of taxes, minus the $100 billion for removal.

• Unauthorized immigrants are not the only domestic group absorbing more in benefits than they pay in taxes, not by a long shot. Leaving aside the elderly (most of whom have paid taxes in the past), Americans with less than a high school degree receive much more in benefits than they pay in taxes because of the progressive nature of our income taxes.

• A big reason this is particularly true today is because the federal government budget is in deficit. The Trump tax cuts have increased deficits, which were already large. Because tax rates are low relative to rates of spending, most living Americans' net fiscal contributions are negative. As a nation we are shifting this burden onto our children, their children, and the children and grandchildren of immigrants.

• The main reason why immigrants are costly to government is because their children go to public school. The data also show that immigrants are actually better than natives in terms of generational mobility in education. These two aspects combine to imply that it is a good deal to educate immigrant children, who pay back those benefits with tax contributions when they are older. Measuring fiscal burden during a single year misses this important benefit.

When parsimony isn't parsimonious

A very cool graphic from a working paper by Larry Bartels quoted in a tweet from Matthew Yglesias:

This Larry Bartels chart contains so much insight. pic.twitter.com/rtm1vaRuGY

— Matthew Yglesias (@mattyglesias) August 20, 2018

only I don't see how it's clear why the "Limited Government" dimension (one of 2 that apparently capture 78% of the observed variance in 43 YouGov survey questions from November 2017) reflects "classic economic conservatism." Table A in the Appendix appears to show the index varies a lot with things that are quite different from "economics," like Fox News and the United Nations. I guess the boldface might denote statistical significance, which if true suggests that Fox News might be insignificantly associated with the index but with a large correlation/coefficient. 

Oster's parenting book due Spring 2019

I'm looking forward to this with great excitement!

If my book were my baby (which, let's face it, it kind of is) today would be the positive pregnancy test...  due date April 23 2019 https://t.co/aOTOCyHsmQ

— Emily Oster (@ProfEmilyOster) August 14, 2018

Gains against maternal mortality in CA

NPR reports on efforts in California by the CMQCC to reduce maternal mortality. There's a lot here, including discussions of the proximate causes: hemorrhage and preeclampsia (hypertension). And the article includes some illustrations of policies that have reduced maternal mortality in the state, like special "hemorrhage carts" and procedures. One of the more interesting points was about referrals from rural hospitals to medical centers when conditions were risky. One imagines that if this is important in California, it's probably also important in Texas and in many other regions where population density can be low.

Oster on breast feeding

The breastfeeding news.

1. Covered in detail in new book!
2. Long term benefits (IQ, obesity, etc) way overstated in developed world; short term also less than claimed.
3. In places without clean water, effects much bigger.
4. Should encourage both BF and safe formula options.

— Emily Oster (@ProfEmilyOster) July 10, 2018

Very nice and brief summary. Cleanliness is no joke, and breast feeding is known to be a lot cleaner and safer for infants than formula with contaminated water or in dirty bottles. But other benefits are far less clear.

Immigrants in the military

Several outlets including NPR were reporting last week about discharges of immigrants seeking to join the military, including the Army Times. It appears the development may have to do with the program for Military Accessions Vital to the National Interest (MAVNI), which emphasized recruiting for language skills especially in the Army, according to this Military.com article. A Vox.com article seems to identify security concerns stretching back to the Obama Administration combined with difficulty verifying backgrounds of immigrants from countries with less robust information systems.

Setting aside the impact on individuals for a moment, it is also useful to assess how big of a challenge this is for the military. The MAVNI document above, which appears to have been written in 2015, states that about 5,000 green card holders enlist each year, and that 109,250 have attained citizenship through military service since 9/11.

Active-duty force strength is about 1.34 million, according to Pew Research. It is harder to find the percent who are foreign born. In 2008, when forces were roughly the same size as they are today, there were about 65,000 immigrants on active duty, or about 5% of all active-duty forces, and more than two thirds had been naturalized. Ten years may have changed this somewhat, but these magnitudes imply something like 20,000 green card recipients or around 1.5% of forces.

Trade wars and recessions

A vox.com article by Matthew Yglesias references a great blog post by Paul Krugman guesstimating the outcome of a global trade war as a tariff hike of 40%, a reduction in imports of 70%, and an associated welfare cost of 2.1% of GDP.

Krugman is right that this isn't enormous, but I wanted to add this perspective:  If the labor share of GDP is 2/3, you would also end up with a 2.1% hit to GDP if you had roughly a 3% reduction in labor inputs. Since the financial crisis of late 2008, U.S. labor force participation has fallen about 3.2%, from 66.0% to about 62.8%.

I think that's revealing: it's consistent with the point that the world likely won't end as a result of a shock of this magnitude; and it's also useful to contextualize the shock this way, as something like the wave that swept a lot of people out of the labor force.

Breast feeding, parental leave, and child outcomes

The latest Administration kerfuffle on the international stage concerns the WHO and breast feeding. It's hard to know exactly what happened. An NPR interviewee describes the issue as having to do with deceptive marketing practices of formula companies; the BBC reports it was about striking language that promoted breast feeding, and inserting the phrase "evidence based."

What does the evidence suggest about breast feeding? I don't think it's as cut and dried as the BBC article pithily states: "years of research has found breast milk to be healthier than other substitutes." While we wait for somebody like Emily Oster to sift through the medical advice for us in her forthcoming book, here are some references I found helpful.

As you might imagine, there are few studies that are able to isolate the effects of breast feeding from family leave, which tends to increase breast feeding.

• Baker and Milligan (JHE, 2008) examine changes in Canadian maternity leave and find relatively little evidence of any effects on children's well-being within the first 24 months.

• Fletcher (Health Econ., 2011) looks at long-term educational outcomes and self-reported (retrospective!) breast feeding in the Add Health panel. He finds that differences in outcomes associated with breast feeding appeared to be driven by differences in parental behavior like treating siblings equally.

• Colen and Ramey (Soc Sci Med, 2014) assess patterns in outcomes among children in the NLSY across a broad age range, 4 to 14, associated with breast feeding. They report that benefits were mostly associated with parental characteristics, which prompted their selection into breast feeding.

• Lichtman-Sadot and Bell (JPAM, 2017) look at the introduction of paid family leave in California in 2004 and compare cohorts of children born before and after the policy change. They report large effects of paid family leave on 6 health metrics observed mostly via parental self-report at Kindergarten.

I would say there is some evidence that family leave can improve outcomes, but there is not much evidence to suggest that the primary channel is breast feeding.

Native employment rose with immigration

A colleague and I examined ACS data from 2005 to 2016 and found evidence that native employment actually rose along with U.S. immigration, contrary to what one might expect based on basic economic theory where labor demand isn't shifting.

Mao-Mei Liu and I applied a standard statistical model in social science, panel fixed effects, to public-use microdata areas (PUMAs) in the U.S. over the period covered by the American Community Survey (ACS). We were concerned that we'd find something that was irregular but only because of the Great Recession. One could easily imagine seeing that the recession dried up employment for immigrants and natives alike, and that would tell us little about how the two themselves were related.  But instead what we found was a remarkably robust result across time periods and Census divisions: native employment seemed to trend upward by a small amount with immigration.

I'm particular keen on the point estimate of the average effect, which showed something like 1 additional native employed for every 20 more immigrants in the local labor force.  This sounds a lot like a situation where foreign-born labor combines with native labor, where the native labor might specialize in communications. There could be many other ways in which labor combines, and we did not examine industries or occupations.

Marshmallows and wealth

Here’s a nifty Atlantic article about the famous Marshmallow Test of delayed gratification in childhood and later-life outcomes. As one might expect, family background matters a lot in determining whether you can wait on that marshmallow. An easy way to see why is to imagine why a child might be hungry for the marshmallow.