Geography (Urban/Rural)

Dan Frosch, Kris Maher and Zusha Elinson, "Rural America Reels From Violent Crime. ‘People Lost Their Ever-Lovin’ Minds.’,"Wall Street Journal, June 10, 2022, https://www.wsj.com/articles/violent-crime-rural-america-homicides-pandemic-increase-11654864251

Rural America Reels From Violent Crime. ‘People Lost Their Ever-Lovin’ Minds.’

Local prosecutor Rebecca McCoy used to think of her home in central Arkansas as a place where the worst crimes were usually stolen tractors and lawn mowers.

In March 2020, she was called to the trailer of a 72-year-old man who had been bludgeoned to death with a baseball bat. It was White County’s first homicide in almost two years. By that December, there were 11 more.

In Marion County, a swampy stretch of South Carolina, Sheriff Brian Wallace and his deputies worked nine killings in 2021, including the execution-style shooting of an 80-year-old retired teacher whose family the sheriff knew personally. It was the highest annual body count he had seen since he joined the small department more than two decades earlier.

For ranchers Bill and Diana Beck, the violence arrived in their mountain community in northwest Montana in April when a friend texted them a picture of blood in the snow. A local bar owner had been shot to death. It was Flathead County’s 12th homicide since the start of 2020, one of the region’s most violent periods in recent memory.

“I was floored that something like that happened a mile down the road,” said Ms. Beck.

Violent crime isn’t just rising in the nation’s cities. Murder rates across the rural U.S. have soared during the pandemic, data show, bringing the kind of extreme violence long associated with major metropolises to America’s smallest communities.

Homicide rates in rural America rose 25% in 2020, according to the Centers for Disease Control and Prevention. It was the largest rural increase since the agency began tracking such data in 1999. The CDC considers counties rural if they are located outside metropolitan areas defined by the federal government.

The rise came close to the 30% spike in homicide rates in metropolitan areas in 2020.

The CDC hasn’t analyzed 2021 homicide data yet. In some rural counties, murder rates remained high last year, while in others they have begun to recede along with Covid, data from local law-enforcement agencies shows.

County sheriffs are trying to hire more deputies. Small-town prosecutors, unaccustomed to handling numerous homicides cases, find themselves overwhelmed with them.

In cities, law enforcement and civic leaders have blamed the increase in violent crime on factors such as police pulling back after racial-justice protests, the proliferation of guns, initiatives to release more criminal suspects without bail and a pandemic pause in gang-violence prevention programs.

In rural counties, where ties between police and locals are often less fraught, officials say the reasons for the rising violence are hard to pinpoint. They speculate that the breakdown of deeply rooted social connections that bind together many small communities, coupled with the stress of the pandemic, played a role.

Pastors point to the suspension of rituals such as in-person church services, town gatherings and everyday exchanges between neighbors. Such interactions can serve as guardrails, helping to prevent conflicts from turning violent. The psychological and financial stress due to isolation and job loss were especially pronounced in remote areas, where social services were limited even before Covid-19 struck, local leaders say.

As the pandemic took hold in the spring of 2020, fights between family members, acquaintances and even strangers escalated more frequently into deadly confrontations, authorities in some rural counties said.

“You saw the shutdown, crime started increasing, people were on edge, mental-health problems increased, more alcohol was sold,” said Brian Heino, sheriff of Flathead County in Montana.

An American Farm Bureau Federation poll taken in December 2020 found that two-thirds of adults in rural America between the ages of 18 and 44 said they experienced more mental-health challenges than in the prior year. According to a CDC survey conducted in June 2020, 12% of responding adults in rural areas reported starting or increasing their use of drugs or alcohol to cope with pandemic-related stress.

The homicides have been challenging for law enforcement to address. Small numbers of deputies must patrol expansive geographic areas, and the killings were often not connected to each other, making it more difficult to target particular criminal groups or devise effective prevention strategies.

“It’s not like you got gang members dealing drugs, and you can get boots on the ground, and undercover buys and prosecutions,” Marion County’s Sheriff Wallace said. “These are really tough to stop.”

Veteran law-enforcement officials said they had never before witnessed the level of violence of the past two years.

“It was like people lost their ever-lovin’ minds,” said Ms. McCoy, the prosecuting attorney in White County, a dry county in central Arkansas with poultry farms and a Christian university.

In Marion County, S.C., which has about 28,700 residents and more than 100 churches, pews used to be packed every Sunday. After services, people would pour into restaurants to eat seafood and Southern cooking.

When Covid struck, some churches stopped in-person services, and restaurants shut down. Walmart started closing early. People no longer stopped to socialize in the grocery store.

Authorities in rural counties said that as the pandemic wore on, troubled individuals became even more isolated, and in some cases, tipped over the edge.

“There are parts of our county that don’t even have internet service,” said Tammy Erwin, a victims’ advocate for the Marion County Sheriff’s Office. “So it’s not like everyone could just jump on Zoom.”

Just before midnight on Oct. 28, 2020, one of Sheriff Wallace’s deputies came upon an SUV with its windows fogged outside the Park & Blow Handi Mart. He could make out a man inside sweating heavily and sitting on something. When he looked closer, he saw the body of a woman, eyes bulging. She had been stabbed multiple times with an ice pick.

Johnny Michael Dillard, 57, was charged with killing Kemethia Spain, 51, whom he had known for a few years. Several people came forward to say Mr. Dillard had talked openly about wanting to know what it was like to kill someone. He was found not competent to stand trial and has been committed to a facility for the mentally ill.

Ms. Spain’s homicide was one of 15 that Marion County deputies responded to over the past two years, equal to the previous 10 years combined. “I’ve never seen these numbers before, nothing even really close to that,” said Sheriff Wallace.

As lockdowns took hold, the sequestering of family members inflamed household tensions. In Montana, domestic-violence incidents reached a record high in 2020, then jumped even higher in 2021, according to state crime data.

In June 2020, Josh Buls, a detective in the sheriff’s office in Flathead County, arrived at a home in Olney to find the bodies of a woman, her 3-year-old daughter and a male friend. All had been stabbed to death. The perpetrator, the woman’s estranged husband, had shot himself in the head before deputies arrived.

The killings shocked not only the 146 residents of the mountain town, but many in the sheriff’s office, which patrols more than 5,000 square miles of rugged territory bordering Canada. One deputy who responded to the 911 call that day decided to leave the profession, said Mr. Buls.

The six homicides in 2020 in this county of 108,000 were a record, as were the 51 statewide, according to state crime analysts.

Flathead County’s Sheriff Heino said rising methamphetamine use and mental-health problems during the pandemic contributed to more erratic behavior and more violent interactions between officers and suspects. Law enforcement in the county was involved in six shootings since May 2020, including four fatal ones. Before then, the last time a Flathead County deputy had killed anyone was 2007.

When the owner of Snow Slip Motel Bar & Cafe near Essex, Mont., was shot dead in April, news spread quickly among those living near the peaks and forests of Glacier National Park. Jim Noffsinger, a 57-year-old construction superintendent and rancher, locked his front doors for the first time.

“I’ve been accused of many things in my life, but being scared was never one of them,” he said. “Such a place of serenity has taken on a dark shadow.”

For Ms. McCoy, the prosecutor from White County, Ark., the unprecedented string of homicides in 2020 began the same month that Covid hit. Over a nightmarish 10 weeks starting that March 4, six people were murdered in the small towns that dot the county of 77,000, about 60 miles northeast of Little Rock.

Ms. McCoy searched for any common thread. Three of the victims were men in their 70s who lived alone, including one who was shot in the face when he opened his front door.

“At first we’re thinking, wow, if you’re an older man living by yourself, you might want to move out of county, because apparently it’s not safe in White County for you,” said Ms. McCoy.

To keep the cases straight, Ms. McCoy took out a white board she hadn’t used in years. As the killings continued, her “murder board” filled up, and she finally had to rewrite the names of victims and suspects and hearing dates in smaller letters to fit. It was the deadliest stretch of her 23-year career.

She said she felt helpless. “I don’t know what to say, besides y’all stop killing each other,” said Ms. McCoy.

With homicide cases mounting, Ms. McCoy told her staff of prosecutors who had little experience with murder cases to focus on their families when the workday ended. For the first time in her career, she said, she had trouble clearing her mind of gruesome crime-scene images when she returned home to her own family each night.

During an interview in her office in Searcy, Ark., as a rescue cat named Catticus maneuvered between murder case files stacked on the floor, Ms. McCoy shed tears recalling the day in June 2020 when she and county detectives arrived at a house in the town of Bald Knob to find the body of a 5-year-old girl whose throat had been slit by her father. They found the father overdosed in his car in a nearby state park a few hours later. Attempts to revive him failed.

Prosecutors in rural areas have struggled to identify motives behind many of the murders. Ms. McCoy concluded that the stress of the pandemic had contributed to disagreements escalating into violence.

That spring, 77-year-old Julius Williams Jr. was killed in what appears to have been a dispute about his lawn. Mr. Williams had lived much of his life in Kensett, a town of about 1,500 with weathered houses, weedy yards and a single main intersection.

Jason Baxter, a 26-year-old resident of a nearby trailer park, asked Mr. Williams if he could mow the older man’s yard, according to court records. Mr. Williams declined.

Days later, police found him lying inside his front door in a pool of blood. Mr. Baxter admitted to police that he had shot Mr. Williams, who was Black, “right in the eye,” according to a transcript of his police interview.

Investigators found rambling notes written by Mr. Baxter, who is white, about protecting white children.

Lee Short, a lawyer appointed by the state to represent three White County murder suspects, including Mr. Baxter, said Mr. Baxter had mental-health issues.

Mr. Williams’s daughter Thunder Nkere said she couldn’t believe he was killed in the tiny town where she grew up. “It’s one of those little country bumpkin places where everyone knows each other,” she said.

Last year, the number of homicides in White County declined to four. Ms. McCoy said she hoped the easing of the pandemic would help continue the decrease in numbers.

Killings in the county haven’t yet fallen to prepandemic levels, and the sheriff’s office has strained to keep up, even after more than doubling the number of detectives. It recently converted a donated ambulance into a crime-scene vehicle. It also bought new equipment, including a machine that can extract cellphone data in hours, compared with waiting months for a state lab to process such data.

In April, Mr. Baxter pleaded guilty to first-degree murder and was sentenced to 40 years in prison. Ms. McCoy wiped away one line on her murder board.

Violent crime isn’t just rising in the nation’s cities. Murder rates across the rural U.S. have soared during the pandemic, data show, bringing the kind of extreme violence long associated with major metropolises to America’s smallest communities. Homicide rates in rural America rose 25% in 2020, according to the Centers for Disease Control and Prevention. It was the largest rural increase since the agency began tracking such data in 1999. The CDC considers counties rural if they are located outside metropolitan areas defined by the federal government. The rise came close to the 30% spike in homicide rates in metropolitan areas in 2020. The CDC hasn’t analyzed 2021 homicide data yet. In some rural counties, murder rates remained high last year, while in others they have begun to recede along with Covid, data from local law-enforcement agencies shows. Veteran law-enforcement officials said they had never before witnessed the level of violence of the past two years. “It was like people lost their ever-lovin’ minds,” said Ms. McCoy, the prosecuting attorney in White County, a dry county in central Arkansas with poultry farms and a Christian university. For Ms. McCoy the unprecedented string of homicides in 2020 began the same month that Covid hit. Over a nightmarish 10 weeks starting that March 4, six people were murdered in the small towns that dot the county of 77,000, about 60 miles northeast of Little Rock.

New York — which is where I happen to live — isn’t just safe relative to other big cities. The data in the above chart are from the Federal Bureau of Investigation and local police departments, but the Centers for Disease Control and Prevention also keep track of homicides as reported on death certificates, and one can sort them by, among other things, whether they befell residents of a county that contains all or part of the biggest city in a large (population of one million or more) metropolitan area, a fringe (suburban) county in such a metropolitan area, or elsewhere. Suburban counties in large metro areas have tended to have the lowest homicide rates, but New York City’s was even lower from 2017 to 2019 and was only slightly higher in 2020, the most recent year for which the CDC has full data. Some causes of death other than homicide seem pretty relevant to assessing safety. Your risk of getting killed on the subway or in a subway station is much higher in New York City than in a place with no subway, for example,but if you add up all risks related to getting to work or school or wherever else it is you need to go, it’s a different story. New Yorkers are only about as third as likely to die in transportation accidents of any kind as Americans are overall. Put homicide and transportation risks together, and New York starts looking like a refuge from the American carnage.All these accountings leave out the September 2001 terrorist attacks. The most dangerous areas are a more hardscrabble lot, although led by an affluent coastal metro (Baltimore-Columbia) where the main city has an awful violent crime problem. The city of Baltimore’s homicide rate was a staggering 58.5 per 100,000 residents in 2021, twice that of Chicago and more than 10 times New York City’s.

Justin Fox, "New York City Is a Lot Safer Than Small-Town America,"Bloomberg, June 7, 2022, https://www.bloomberg.com/opinion/articles/2022-06-07/is-new-york-city-more-dangerous-than-rural-america

New York City Is a Lot Safer Than Small-Town America

Since the summer of 2020 the United States has been experiencing a deadly wave of violence, and New York City has not been immune. The number of homicides in the city shot up 47% in 2020, and rose another 4% last year. While the tally is down 9% so far this year there have been some shocking murders, not to mention a mass shooting on the subway that miraculously didn’t take any lives.

Regular consumers of the news would thus be forgiven for believing that the Big Apple has become a terribly dangerous place. And yes, it is markedly more dangerous than it was before the pandemic. I’m not here to dismiss concerns about rising crime in the city. But a little context is in order. And by “a little context,” I mean a lot of charts.

The historical context, which a fair number of people are aware of, is that New York’s pandemic murder wave followed a long decline, with the city’s homicide rate in 2021 still less than a fifth what it was in 1990. The geographical context, which seems to be less widely understood, is that while homicide rates fell sharply all over the the US in the 1990s, that decline slowed in the 2000s and reversed starting in 2015. In Philadelphia the homicide rate is now worse than it was in the early 1990s, and in Chicago it’s close. Which means there’s a growing gap between New York City and most of the rest of urban America.

Some causes of death other than homicide seem pretty relevant to assessing safety. Your risk of getting killed on the subway or in a subway station is much higher in New York City than in a place with no subway, for example, but if you add up all risks related to getting to work or school or wherever else it is you need to go, it’s a different story. New Yorkers are only about as third as likely to die in transportation accidents of any kind as Americans are overall. Put homicide and transportation risks together, and New York starts looking like a refuge from the American carnage.

There are other accidental threats listed in ICD-10 that seem less salient in the city than in the country, and in some cases, the suburbs. To name a few: “contact with powered lawnmower,” “drowning and submersion while in swimming pool,” “contact with venomous snakes and lizards,” “victim of volcanic eruption.” So I ran the numbers again for mortality from all external causes.

The explosion of drug-overdose deaths in the US over the past decade is one of the most depressing trends one encounters in CDC Wonder, but it does take us somewhat afield from what people usually have in mind when they consider whether a place is safe or dangerous. Same goes for suicides, which have also risen but not as sharply. Then there are external causes such as falls and surgical complications that are usually fatal only for the very old. If included in the calculations, they tend to make retirement communities look terribly dangerous, which doesn’t seem right.

Here’s what mortality from external causes looks like if one excludes (1) accidental poisoning (how drug overdoses are classified), (2) intentional self-harm, (3) falls, (4) complications of medical and surgical care and (5) sequelae of external causes of morbidity and mortality (after-effects, that is, which are also fatal mostly for the elderly).

This is so opposite from conventional wisdom that it’s worth teasing out why. One obvious explanation is that murder scares people a lot more than accidents do, and homicide risk is in fact highest in urban counties — just not in New York City. Another is that, as a colleague put it to me, you may be less likely to be victim of a crime in New York City, but you’re more likely to witness one. In dense urban environments you simply see a lot more of what’s going on, for good and ill. It’s also possible that the issue is not so much urban versus rural as it is who lives where. An affluent rural or exurban county where lots of people can work from home might be lower risk than a poorer urban one where most can’t. Rural counties in Appalachia or the Deep South might have very different profiles from those in New England or the Plains states.

A look at 2020 selected-external-causes mortality by state offers some support for the latter interpretation. The Northeast is the safest part of the country, but some heavily rural states (Mississippi, Arkansas) are a lot more dangerous than others (Vermont, Iowa).

But let’s not forget those data limitations. While CDC Wonder knows everything about mortality in the US, for privacy reasons it only shares what it knows if there are 10 or more deaths to report, and because of reliability concerns it only calculates mortality rates if there are 20 or more deaths. For the collection of external causes I selected, Wonder churned out 2020 death totals for 1,711 of the nation’s 3,143 counties and county equivalents (parishes in Louisiana, census areas in Alaska, independent cities in Virginia and some other states) and mortality rates for 997. It provided enough data to calculate the mortality rates for all 1,711, but I chose not to for the charts here. If I had, three more Northern Virginia cities and counties would have entered the top 10 along with Story County, Iowa, and La Crosse County, Wisconsin.

If I selected a wider array of external causes, or combined data from several years, more counties with smaller populations would enter the mix. Still, pretty much any version of the most-dangerous-counties list would be at least as heavy on rural counties as this.

Also, because complete 2020 Census results have yet to be released, the CDC is still using outdated 2020 population estimates based on the 2010 census. For New York City, this estimate is 550,977 lower than the number from the 2020 census, and the CDC estimates for most of the 2010s are probably too low as well. To give a sense of the impact, the city’s CDC-calculated homicide rate of 5.4 per 100,000 in 2020 would drop to 5.1 using the census population. I decided to stick with the CDC’s numbers nonetheless because (1) the city likely lost several hundred thousand residents in the months following the April 1 census record date, and (2) redoing all the 2020 mortality-rate calculations using census data would have introduced a lot of opportunities for error. But keep in mind that all the mortality rates for New York City in recent years in the CDC-derived charts are probably slight overestimates.

Finally, it would be nice to see this data by metropolitan area, wouldn’t it? CDC Wonder’s most up-to-date deaths databases does not provide statistics by metro area. It is possible to download them by counties, then sort those into metropolitan areas, but with 2020 deaths data from my list of external causes available for only a little over half of US counties, that doesn’t really work. By combining deaths from the five most recent available years I was able to get a more complete county list, and fill in the blanks with a little extra effort. Here, then, are the safest metropolitan areas in the US from 2016 through 2020, by my admittedly somewhat idiosyncratic definition.

The most dangerous areas are a more hardscrabble lot, although led by an affluent coastal metro (Baltimore-Columbia) where the main city has an awful violent crime problem. The city of Baltimore’s homicide rate was a staggering 58.5 per 100,000 residents in 2021, twice that of Chicago and more than 10 times New York City’s.

Data sources for charts

Safest of the Six Largest U.S. Cities: The main data sources here are the FBI’s Crime Data Explorer and Census Bureau annual population estimates that I got from the Data Commons Place Explorer (for pre-2010 numbers) and the Census Bureau (for 2010-2021). I could not find city population estimates for 1985-1989 so I calculated my own by just assuming the cities grew (or shrank) at a uniform pace between the 1980 and 1990 censuses. The FBI hasn’t released 2021 crime data yet so I got that year’s numbers either directly from police departments or from news reports based on police department data. Also, for some reason the FBI’s post-2014 homicide numbers for Philadelphia are different from (and lower than) those on the Philadelphia Police Department website, so I went with the latter.

Murder in the Cities, Suburbs and Country: This is downloaded straight from CDC Wonder’s Underlying Causes of Death database, but there are two different ways to get homicide totals from that. One is to look under “Injury Intent and Mechanism” and select “homicide” as the injury intent, the other is to look through the ICD-10 codes used to classify all causes of death and select “assault” under “External causes of morbidity and mortality.” The former includes deaths from terrorist attacks and the latter does not, because they are classified under a separate ICD-10 category of “Codes for special purposes.” When this piece was first published, the chart showed homicide deaths with the September 2001 terrorist attacks subtracted out (for consistency with the FBI data and because otherwise New York’s high 2001 rate would make the chart very hard to read); subsequently I realized it would be simpler and more consistent with the other charts to use the assault numbers, which are almost but not quite identical. Here they are by county urbanization status (I combined the two categories of nonmetro counties for the charts) and for New York City. Once you click through, you’ll have to agree to a couple of data use restrictions, then wait as CDC Wonder generates the data.

Car Crashes Change the Picture: This adds together assault plus transportation accident deaths as reported under “External causes of morbidity and mortality.” Here they are by county urbanization, and for New York City.

Safer in the Big City: This is all “External causes of morbidity and mortality,” by county urbanization, and for New York City.

Still Safer in the Big City: This is external causes minus the categories discussed in the text and the chart footnotes, by county urbanization and for New York City.

Safe States and Dangerous Ones: Selected external causes by state for 2020.

The Safest Counties in America, Sort Of, and The Most Dangerous Counties in America, Sort Of: Selected external causes by county for 2020.

Evidence that NAFTA drove working class voters out of democratic party to GOP, “…We begin by showing thatNAFTA had a major, negative effect on employment in areas exposed to Mexican import competition. By 2000, counties in the top quartile of our measure of NAFTA exposure saw a 5-8 log point decline in total employment, relative to the bottom quartile. These losses were concentrated in manufacturing and, importantly, exhibit no pre-trends from the mid 1980s to 1994. While we begin all of our analysis by showing trends with county year raw data, the basic shape of our event-study coefficients are unchanged as we add a large number of controls: pre-period county-level measures (e.g., 1990 manufacturing share of employment, 1990 share with a college degree) interacted with year fixed effects, to control flexibly for other secular changes (e.g., automation, skill biased technological change) that may affect communities differentially across time; the “China shock” measure from Autor et al. (2013) interacted with year fixed effects, to ensure we isolate the NAFTA effect from the rise of Chinese imports; and fixed effects at the state×year level, to pick up any policy or other unobserved variation within states across time... Having documented large, negative local employment effects in communities exposed to Mexican import competition, we then ask how voters in these communities reacted. NAFTA was a major issue in the 1992 U.S. presidential campaign, with Ross Perot making opposition to it a major motivation for his third-party campaign, a third-party campaign that picked up 19 percent of the popular vote, making it the most successful such campaign since Teddy Roosevelt’s run as the Bull Moose Party candidate in 1912....We focus on House election votes in most of our political analysis, as every House seat is up for election every two years, allowing for a balanced panel of years. While NAFTAexposed counties (many in the upper South) begin the sample period more Democratic as measured by House election votes than the rest of the country, they exhibit a sharp change in trend and become increasingly Republican. In contrast to our employment effects, which show no pre-1994 pre-trend regardless of specification, this political turning point occurs in either the 1992 or the 1994 election, depending on the controls we include in the event study analysis. We find this ambiguity unsurprising, given the political salience of NAFTA in the 1992 election, even though its provisions did not go into effect until January 1994. Beyond the ambiguity between 1992 and 1994, there is no political pre-trend in NAFTA exposed counties from 1980 to 1990. The shift we document is large. While these counties are in 1990 the most Democratic of our four quartile groups, by 2000 they are as or more Republican than any of the quartiles. We present a variety of microdata-based evidence that the political shift was indeed due to NAFTA. First, we show that in the areas we define as most vulnerable to NAFTA, survey respondents significantly oppose NAFTA and this opposition continues to the present day. Second, in repeated cross-section data from the American National Election Surveys (ANES) we show that, in each year of survey data from 1986 to 1992, Democrats enjoy a significant and steady advantage among those with protectionist views, but between 1992 and 1996 a significant number of protectionist voters move toward the GOP and remain there. Finally, in an ANES panel dataset from 1992 to 1994, we can look at the same voters over time during this key moment. We indeed find that a significant share of those who in 1992 express protectionist views have moved their party-identification toward the GOP by 1994. We show these effects are robust to flexibly controlling for a variety of demographic variables as well as views on other political and policy questions...This shift among protectionist respondents is larger for whites (especially men and those without a college degree) and those with conservative social views, suggesting an interactive effect whereby racial identity and social-issue positions mediate reactions to economic policies…"

Jiwon Choi, Ilyana Kuziemko, Ebonya Washington and Gavin Wright, “Local Economic and Political Effects of Trade Deals: Evidence from NAFTA,” National Bureau Of Economic Research, November 2021, https://www.nber.org/papers/w29525

“…We show the results of estimating the event-study equation (5) in Figure 13. The first series shows the coefficient estimates from an equation with no controls (so the vector Xi is omitted), so the plotted points are just raw differences. We see that pre-NAFTA, those with protectionist views were less supportive of the GOP. Sometime between 1992 and 1996 (the protectionist question is not asked in 1994) a significant number of protectionist voters moved toward the GOP, so that the raw difference disappears. In Appendix Figure?? we show this result as two separate series instead of as a difference, and we also include 2016, which shows that in that year there was yet again another movement of protectionist voters toward the GOP, consistent with Trump's anti-trade message. In the second series, we add standard demographic and socioeconomic controls: race, sex, age, education, and income. The same 1992-1996 shift toward the GOP among protectionist voters remains, even after allowing these characteristics to have their own effect each year. In the final series, we add controls for trust in the federal government, views (a thermometer" going from cold to warm) toward African Americans (given the importance of race in U.S. politics), views on abortion, and weekly religious attendance. We are limited somewhat by wanting to use questions that are asked each or most years in our sample. For example, questions on gay rights and regulations on _rearms purchases are missing in our pre-period (we are able to add a richer set of controls in our panel-data analysis in the next subsection). As the final series of the figure shows, the large shift between 1992 and 1996 remains even after flexibility controlling for views toward the other salient issues available over our sample period….”

“…Figure 12 makes clear that, throughout the 1980s, NAFTA-vulnerable counties exhibited a steady Democratic lean in House elections, but beginning in 1994 shift in the Republican direction….”

Evidence, “…Our first set of political results takes the same county-level event-study approach we have used for our local-economic analysis, except that our dependent variable occurs every other year, as we focus on House election outcomes. Because of the important role of NAFTA in the 2016 election, we extend our post-period to 2016, though readers are free to discard evidence this far from the original 1994 event. Except for the extension of the sample period and our observing the dependent variable biennially instead of annually, the analysis remains identical to those in the previous sections. Figure 12 shows the House election outcomes using our standard event-study specifications. Unlike the employment and other economic outcomes, which had a clear break beginning in 1994 (the year of NAFTA implementation), our baseline specification shows a break beginning in the 1992 election. Given the large role that the debate over whether to ratify NAFTA played in the 1992 election, it is not clear that 1994 as opposed to 1992 is the most legitimate start of the post-period" for the political analysis. Adding state-year fixed effects generally does not change the event-study coefficients, though now 1994 appears more like the break in trend than does 1992. It appears that within states, NAFTA-exposed counties break from their pre-period trend in 1994, even if the break occurs in 1992 when both within- and between-state variation in NAFTA exposure is considered. As we have found throughout the previous analysis, adding the ADH CZ-level measure interactions with year fixed effects (the second series in Figure 12) makes no appreciable difference to the event-study coefficients. Nor does 1990 county-level manufacturing interacted with each year. Again, as with previous outcomes, the set of covariates that most attenuates the implied effects of NAFTA are the 1990 college share interacted with year fixed effects. But even with this specification, we find a visually compelling shift in the two-party Democratic share of vote in House races. In this specification, there is a at pre-trend with no coefficient’s confidence interval exclude zero, but in the post-period an average coefficient value of roughly -0.75 (compared to the other specifications, where the post-period coefficient values are more like -1.5). What do these magnitudes suggest in terms of actual political outcomes? Taking our preferred specification (the second series, with state_year _fixed effects) the average of the coefficients in the late 1990s and early 2000s is roughly 0.60, suggesting that the most exposed quartile increase its share of GOP House votes by roughly 4.8 percentage points relative to the least exposed. As Table?? shows, the baseline for this group was only 0.38, so the 4.8 percentage point swing is very large. Appendix Figure A.12, which presents the data in completely raw form, shows that by 1998, these once-solidly Democratic counties voted as or more Republican in House elections as did any other county quartile…”

The political response in areas vulnerable to NAFTA

Evidence, “…Figure 4 shows our four groups of counties, with data unadjusted except for each group's (log) employment level being normalized to zero in 1993. While in the event-study analysis we weight by 1990 county population (both to match past work as well as to better match to the political results, where counties are essentially weighted by population of voters) in this raw analysis we do not. The four series are nearly on top of one another in the years leading up to NAFTA. But they begin to diverge thereafter. By 2000, the most-exposed quartile has roughly five to ten log-points less than other counties relative to the 1994 baseline. Note also that the second-most-exposed quartile also shows a slowing of employment growth relative to the least-exposed half of counties, suggesting that the effect is not entirely driven by a few counties in the most-exposed quartile… As shown in Table A, counties that would be more exposed to NAFTA were already different on important dimensions in 1990: for example, they had higher reliance on manufacturing employment and lower rates of college-degree completion. The fourth series in Figure 5 adds 1990 manufacturing share of county employment interacted with year fixed effects, which barely moved the coefficients. The final series is the same, but instead adds 1990 share of adults with a college degree interacted with year fixed effects. Of all the controls we add, this one has the most appreciable effect. Nonetheless, the effects are large, negative and highly significant..."

China). While growing before NAFTA, Mexican imports enjoy more rapid growth beginning in 1994. Interestingly, despite the larger focus on China in the empirical labor economics literature, it is not until 2004 that China supplants Mexico as the most important low income source of imports (though its rise since 2004 is indeed more rapid than any period for Mexico)….”

Good factoid, “…Figure 1 shows the value of imports to the US from Mexico (and, for context, also includes

Xavier Giroud, Simone Lenzu, Quinn Maingi and Holger Mueller report on their research (attached) that showed productivity spillovers diffuse through the economy though firms with dispersed geographic footprints, "....We first assess the effects of an increase in within-firm, across-location knowledge sharing on the distribution of economic activity. Intuitively, an increase in knowledge sharing between plants of multi-region (‘multi-county’, or MC) firms makes these plants more productive. Given a reduction in marginal costs, the plants increase labour demand, pushing up nominal wages (but not enough to offset the reduction in marginal costs). As a result, goods prices decline, which pushes up real wages. At the macroeconomic level, this influences the spatial distribution of economic activity. Surprisingly, knowledge sharing between plants widens economic disparities across regions. In relative terms, highly populous regions with high real GDP become richer, while rural regions with low real GDP become poorer (Figure 1, panel A).The fundamental economic force driving this result is the sorting across locations by multi-county plants. In equilibrium, counties with more multi-county plants experience relatively large average real wage increases, as well as labour inflows, while counties with fewer multi-county plants experience smaller average real wage increases and thus labour outflows. In our model economy, as well as in the data, 60% of all multi-county plants are in the 10% most populous counties (Figure 1, panel B). Hence, the productivity gains from an increase in knowledge sharing between plants are disproportionately concentrated in the richest, most populous counties....we examine the welfare implications of place-based policies in environments with and without knowledge sharing between plants. In each scenario, we simulate local productivity shocks - such as those arising from place-based policies - and measure the resulting changes in worker-level utility. While, on average, the welfare gains are much higher with knowledge sharing (64%), they are unevenly distributed. In particular, workers in other regions working for plants that are connected (through plant-level networks) to the region with the productivity shock experience utility gains that are five times larger than workers at non-connected plants. At the regional level, the welfare gains are disproportionately concentrated in highly populous regions, as those tend to have the most connected plants. Hence, knowledge sharing through plant-level networks leads to a highly uneven distribution of the welfare gains from local productivity shocks. These results complement the ones in Gaubert (2018a, 2018b), proving an additional explanation for why place-based policies may increase spatial disparities across regions....."

Xavier Giroud, Simone Lenzu, Quinn Maingi, Holger Mueller, "Propagation and amplification of local productivity spillovers," Center For Economic And Policy Research, December 1, 2021, https://voxeu.org/article/propagation-and-amplification-local-productivity-spillovers

Propagation and amplification of local productivity spillovers

Local governments spend billions of dollars every year on subsidies for companies to locate within their jurisdictions (Story 2012, Bartik 2020). From an efficiency standpoint, a common rationale for such place-based incentives is the existence of ‘Marshallian agglomeration economies’ - spillovers to the local economy in the form of input sharing, labour market pooling, and knowledge externalities - that raise the productivity of local workers and businesses (Moretti 2010, Kline and Moretti 2014, Neumark and Simpson 2015, Gaubert 2018a, 2018b). While the success of each individual place-based policy ultimately depends on how the policy affects the equilibrium balance of agglomeration and congestion forces (Redding and Esteban 2016), it is widely believed that productivity gains are highly localised and confined to the region in which the policy is implemented. In a recent paper (Giroud et al. 2021), we challenge this view, providing both reduced-form and model-based evidence showing that local productivity spillovers may propagate throughout the economy through the plant-level networks of multi-region firms.

Local versus global productivity spillovers

To identify local productivity spillovers, our empirical analysis builds on the natural experiments in Greenstone et al (2010), who study the effects of large plant openings (‘million dollar plants’, or MDPs) on the productivity of incumbent plants. In this setting, plants in ‘runner-up’ counties, which narrowly lost the competition to attract an MDP, provide a counterfactual for plants in the ‘winner’ county, where the plant was ultimately located.

Using confidential plant-level data from the US Census Bureau, we show that the MDP openings raise the productivity of incumbent plants by 4%. This local productivity spillover is strong within a 50-mile radius around the plant, weaker within a 100-mile radius, and insignificant beyond. Hence, and consistent with a large empirical literature, the productivity spillovers between (plants of) different firms decay rapidly with geographical distance.

A different picture emerges when one examines how productivity spillovers spread spatially within firm boundaries. We consider large multi-plant, multi-region firms which are exposed to the MDP openings (and thus to the local productivity spillover), by having a plant in the winner county. Comparing plants in the same county, industry, and year that belong to firms which either own plants in the winner county (‘treated plants’) or in the runner-up counties (control), our research finds that in response to the MDP openings, treated plants outside the winner county experience productivity gains of 1.8%, along with employment gains of 1.6%. Unlike the local productivity spillover, which takes place across plants of different firms, this global productivity spillover, which takes place across different plants of the same firm, does not decay with geographical distance. Indeed, the estimated global spillover effect remains virtually unchanged if we exclude all plants within a 500-mile radius, or within the same state or census division as the MDP in question.

In A survey article, Rosenthal and Strange (2020) write:

“Implicit in the idea that spatial concentration increases productivity is another idea: the degree of proximity matters. Agglomeration economies must decay with distance. How close, then, do firms and workers need to be to each other to benefit from agglomeration economies? Or more colloquially, how close is close?”

We provide a nuanced answer to this question. On the one hand, firms must have a nearby plant to benefit from local knowledge spillovers, as the local agglomeration economy is strongly significant only within a 50-mile radius around the MDP. On the other hand, not all of a firm’s plants need to be located nearby. In fact, it may suffice if only one of the firm’s plants is near the MDP. Once the knowledge spills over to that plant, it can be passed on to other plants inside the firm’s boundaries, increasing the productivity of distant plants hundreds of miles away.

Knowledge sharing across regions within firm boundaries

What explains the global productivity spillover? The evidence presented in our latest paper is consistent with knowledge sharing between plants of the same firm across different regions. While the local productivity spillover may be due to either labour market pooling or knowledge externalities, it is unlikely that a thicker labour market in the winner county would affect the productivity of treated plants hundreds of miles away. Knowledge, on the other hand, can be used in local and distant plants alike. Indeed, once it spills over to the firm’s local plant, it can be freely shared with other plants inside the firm’s boundaries (Markusen 1984). Consistent with this notion, the global productivity spillover does not decay with geographical distance. Further, and consistent with knowledge sharing, the global productivity spillover is much stronger if the distant plant and either the MDP or the firm’s plant in the winner county are in the same industry or in knowledge-based industries characterised by mutual R&D flows or patent citations.

Does knowledge sharing between plants reduce regional disparities?

While the reduced-form evidence points to sizeable spillover effects on distant regions, the importance of the significance of plant-level networks for regional and aggregate outcomes requires a meaningful quantitative framework. Since all plants in the economy are affected through general equilibrium effects (labour flows, goods trade), non-treated plants in distant counties provide an imperfect counterfactual to quantify the significance of knowledge sharing for the amplification of local productivity shocks. For example, as (manufacturing) goods are tradable, changes in goods prices affect real wages across all plants in the economy, including non-treated plants.

To quantify the significance of plant-level networks for the propagation and amplification of local productivity shocks, we develop and estimate a quantitative spatial model with goods trade, labour mobility, plant-level networks, and a rich geography - where plants of the same firm, across regions, are linked through shared knowledge. We estimate the model via indirect inference by exploiting the close link between structural parameters and reduced-form regression coefficients, and then use the structure of the model to undertake counterfactual analyses.

We first assess the effects of an increase in within-firm, across-location knowledge sharing on the distribution of economic activity. Intuitively, an increase in knowledge sharing between plants of multi-region (‘multi-county’, or MC) firms makes these plants more productive. Given a reduction in marginal costs, the plants increase labour demand, pushing up nominal wages (but not enough to offset the reduction in marginal costs). As a result, goods prices decline, which pushes up real wages. At the macroeconomic level, this influences the spatial distribution of economic activity. Surprisingly, knowledge sharing between plants widens economic disparities across regions. In relative terms, highly populous regions with high real GDP become richer, while rural regions with low real GDP become poorer (Figure 1, panel A).

In our second counterfactual, we examine the welfare implications of place-based policies in environments with and without knowledge sharing between plants. In each scenario, we simulate local productivity shocks - such as those arising from place-based policies - and measure the resulting changes in worker-level utility. While, on average, the welfare gains are much higher with knowledge sharing (64%), they are unevenly distributed. In particular, workers in other regions working for plants that are connected (through plant-level networks) to the region with the productivity shock experience utility gains that are five times larger than workers at non-connected plants. At the regional level, the welfare gains are disproportionately concentrated in highly populous regions, as those tend to have the most connected plants. Hence, knowledge sharing through plant-level networks leads to a highly uneven distribution of the welfare gains from local productivity shocks. These results complement the ones in Gaubert (2018a, 2018b), proving an additional explanation for why place-based policies may increase spatial disparities across regions.