Tuesday, 17 November 2026

How the typewriter changed women's work, marriage, and moetherhood

In Western countries, one of the overwhelming trends of the 20th Century was the dramatic increase in women's formal labour force participation. The drivers of that change have been both social and economic, but the role of technological change may be underappreciated. A new working paper by Myera Rashid (Northwestern University) focuses on the changes arising from the adoption of the typewriter. While that might not seem like an important technology to drive meaningful change, Rashid notes that:

Offices in the US adapted to the arrival of the commercial typewriter in the 1880s by shifting towards the available supply of educated women who provided cheaper labor: while women initially made up only 12 percent of the secretarial and typing occupation in 1880, by 1900 they accounted for 75 percent of all typists.

Rashid then looks at how the diffusion of the typewriter across counties affects women's labour force participation, marriage rates, and fertility. To estimate causal impacts, she makes use of a shift-share instrument that combines local industry employment shares measured in 1880 (prior to the adoption of the typewriter) with national shifts in the demand for typists between 1880 and 1920. The labour force participation, marriage, and fertility outcomes are all measured in 1920, and the key data she uses come from Census microdata. In terms of labour force participation, Rashid finds that:

...a 1 percentage point (pp) increase in the predicted change in the number of typists between 1880 and 1920 led to a 0.75 pp increase in share of women in the labor force in 1920. Using a back-of-the-envelope calculation, this estimate implies that without the shock of the typewriter, women’s labor force participation rate in 1920 would have been 3.65 pp lower, equivalent to nearly 1.16 million fewer women in the labor force out of a population of 31.7 million women of working age.

Next, Rashid shows that the effects were not just limited to White women, but there were increases in labour force participation among Black women as well. However, how those labour force changes manifested were quite different between White and Black women. Specifically:

...a 1 pp greater increase in the predicted change in typists had a 1.7 pp higher share of Black women in the labor force in 1920. I show that the main mechanism underlying this change is the reallocation of White women out of unpaid housework and into clerical employment, which increased the demand for Black women’s labor in paid household services work.

Turning to impacts on marriage and motherhood, Rashid finds that:

The labor market effects induced by the introduction of the typewriter further led to a 0.4 pp reduction in the share of women who were married and a 0.6 pp decrease in the share of women who had children in 1920. These estimates imply that the share of married women and women with children would have been 2.04 pp and 2.89 pp higher in 1920, equivalent to 650,000 additional married women and 920,000 additional women with children in the absence of the typewriter.

Importantly, these results relate to the proportion of women with children in 1920, rather than their completed fertility (the number of children they have over their lifetimes). The results could therefore reflect women delaying having children, rather than necessarily having fewer children overall. Nevertheless, what Rashid finds are large effects, and she shows that they are robust to various alternative choices or data sources. I particularly appreciated the use of corresponding data from Canada used to construct an alternative shift-share instrument, which alleviates concerns that there are unobserved time-variant changes at the county level that correlate with the shift-share instrument and might have been associated with the outcome variables. 

The next part of the paper is particularly interesting. Rashid extends her analysis to look at social mobility, specifically, she tracks young white women from the 1920 to 1940 Census, looking at how the ranking of their husbands within the income distribution (in 1940) are related to the ranking of their fathers within the income distribution (in 1920), and how that relationship differs for women who are typists from other white women. She finds that:

For women who grew up at the 25th percentile of their father’s income distribution, working as a typist is associated with marrying men with an average income rank that was 18.4 percentiles above the average rank of husbands of non-typists. Importantly, this mobility advantage is not present in alternative white-collar occupations such as teaching, despite the fact that these occupations offered women similar wages.

Using the same shift-share instrument (SSIV) as for the previous analyses, Rashid then finds that:

...conditional on father’s socioeconomic status, women living in counties with a one percentage point greater increase in the predicted change in typists married husbands whose income rank was 1.27 percentiles higher. Among women with fathers at the 25th percentile, residing in a county with an above median value of the SSIV is associated with marrying men with an average income rank that was 4 percentiles higher

Importantly, again these results are robust to alternative methods and data. Importantly, Rashid also compares sisters within the same family, finding that the social mobility advantage of being a typist remains. The mechanism argued to underlie these results is that women working as typists were more able to work alongside men of higher income earning potential than women with a similar education level in other occupations. Rashid specifically contrasts the results with those for teachers, nurses, and telephone operators, finding that none of those other occupations offered the same social mobility advantage than that of typists.

Finally, Rashid draws some parallels in the conclusion between the technological changes brought on by the diffusion of the typewriter and those arising from generative AI. She notes that while there may be displacement effects as a result of new technology, there are also increased labour market opportunities, spillovers between different groups of workers, and social consequences as well. While it's unlikely that generative AI will have a similar effect on marriage and fertility given the baseline conditions are very different now than they were in the late 19th Century, it is difficult to anticipate what, if any, social changes may arise due to the diffusion of this new technology. The history of the typewriter is a useful reminder that the consequences of technological change can extend far beyond the workplace.

Incidentally, Rashid is on the economics job market this year, and according to her website this paper is currently at the second-round revise-and-resubmit stage at the prestigious American Economic Review.

[HT: Marginal Revolution]

Sunday, 15 November 2026

This week in research #152

This week I attended the 73rd North American meetings of the Regional Science Association International (aka the NARSC Conference) in Philadelphia. Philly is a great city to visit, and the conference was excellent. My own presentation was a reprisal of my ERSA Conference paper on the increasing trend towards natural population decrease (more deaths than births) across the territorial authorities and local boards in New Zealand, which will become an increasingly serious policy issue in the future. Here are some of the highlights I found from the conference:

  • Steven Bond-Smith (session 8 in the programme) recognised that working from home has a different geographical distribution from AI exposure, and used that to disentangle the effects of working from home and generative AI on junior jobs, finding that junior jobs have fallen most in larger cities as a result of generative AI
  • Peter Batey (session 35 in the programme) gave an excellent talk about the legacy of Charles Booth, the originator of the social survey, and his maps of London poverty, which represented an application of geodemographics long before geodemographics as a field was even invented
  • John Winters (session 38 in the programme) used data based on women's birth county to overcome the endogenous sorting by location and disentangle fertility trends between metro and non-metro areas in the US, finding that housing costs largely explain the different trends in fertility between metro and non-metro areas
  • Chiara Del Bo (also session 38 in the programme) described the size and distribution of the 'space economy' in Italy, a first step towards developing a satellite account for the space economy
  • Ephraim Nkonya (session 46 in the programme, which was also the session I presented in) showed that US birthrates have declined four times faster in rural than in metro counties over the period from 2019-2023, and that crude birth rates are higher in metro than non-metro areas 
  • Luyi Han (session 59 in the programme) found that broadband internet is positively associated with working from home among non-metro women, but not among non-metro men
Aside from the conference, here's what caught my eye in research over the past week:

  • Chua and Tsiaplias (open access) find that the COVID-19 lockdown in Melbourne had a statistically significant and economically meaningful disinflationary effect on grocery prices
  • Chen et al. find that hosting the Olympics is positively associated with bilateral trade, using a gravity model covering 187 countries from 1950 to 2019
  • Cuesta et al. (open access) develop a theoretical framework for human rights economics

Saturday, 14 November 2026

Try this: Platypus Economics

One of my favourite economists, Justin Wolfers, has started a new venture: Platypus Economics. He says:

Television tells stories but doesn't explain. A ninety-minute lecture explains but doesn't travel. I'd like to build something in between. Work that makes you a little smarter for next time, not just this one.

Actually, it's not really new - it dates back to April. I only just found out about it this week, even though it received coverage in the New York Times in May. Through the site, Wolfers provides links to a YouTube Channel, which has tons of content. This includes series entitled "Diving In", which is a bunch of economics explainers, "Off the Clock", which unpacks the week's economics news, and "The Professor is In", which responds to audience comments and questions. The most recent video (as of yesterday) is entitled "The Most Expensive Shag in the History of Humanity", about the Jensen Huang scandal.

Platypus Economics also has an associated Substack newsletter, and a podcast that is available on all your favourite podcast apps.

Even better, the site has a bunch of Stata worksheets that you can use to replicate the analyses from many of the explainer videos. That makes it an invaluable resource for students (at least, for students who have access to Stata!).

There is so much content to get into. I have barely scratched the surface myself. Enjoy!

[HT: Stata News]

Wednesday, 11 November 2026

How much did slavery cost the American economy?

What was the economic cost of slavery in the United States? Answering that question requires first recognising that the spatial distribution of economic activity could have been very different if there was no slavery. And accounting for that difference is the approach adopted in this 2025 NBER Working Paper (ungated version here) by Treb Allen (Dartmouth College), Winston Chen (Yale University), and Suresh Naidu (Columbia University).

Using microdata from the 1860 US Census, Allen et al. first establish three stylised facts (with several elements to each):

Stylized Fact #1(a): Free workers sort into higher productivity locations, enslaved workers are sorted even more so into higher productivity locations...

Stylized Fact #1(b): Free workers sort into higher amenity locations, enslaved workers are sorted into lower amenity locations...

Stylized Fact #1(c): Free workers sort into locations with less coercion, enslaved workers are sorted into places with more coercion...

Stylized Fact #2(a): There is no correlation between cotton suitability and agricultural employment share for either enslaved or free workers...

Stylized Fact #2(b): Free workers sort out of agriculture in high malaria areas, whereas enslaved workers are sorted into agriculture in high malaria areas...

Stylized Fact #2(c): Both free and enslaved workers are more likely to be employed in agriculture the greater the distance to freedom, but this is especially true for enslaved workers...

Stylized Fact #3(a): Both free workers’ wages and the price of enslaved workers are higher in locations with higher productivities...

Stylized Fact #3(b): free workers’ wages are lower in locations with better amenities, whereas the price of enslaved workers are higher in locations with better amenities...

Stylized Fact #3(c): free workers’ wages are lower in locations with greater coercion, whereas the price of enslaved workers is (slightly) higher in locations with greater coercion.

The first stylised fact shows that enslaved and free workers are geographically sorted differently. While both enslaved and free workers locate more in high-productivity areas, this is for different reasons. Free workers can capture higher wages in those areas, while enslaved workers are more profitable for their owners in those areas. Allen et al. use malaria suitability as a negative indicator of amenity. Free workers avoid areas that are more susceptible to malaria (low amenity), whereas enslaved workers have no choice and cannot avoid those areas. Allen et al. measure the extent of coercion with the distance to the North, since slaves who were located further from the North could be more heavily coerced in work without prompting them to attempt to escape to the North. Unsurprisingly, there are more slaves further from the North, where coercion is easier for slave owners.

The second stylised fact shows that enslaved and free workers are sorted into different sectors of the economy, with enslaved workers more likely to be engaged in agriculture, especially in low amenity (high malaria) areas and areas with more coercion (further from the North). This sorting doesn't simply reflect the suitability of areas for agriculture (specifically, suitability for cotton).

The third stylised fact shows that wages are higher where productivity is higher. Lower wages for free workers in high amenity areas reflects a compensating differential. That is, wages need to be higher in low amenity areas in order to attract free workers. Where coercion is greater (further from the North), free workers earn lower wages as they need to compete with more enslaved workers.

After establishing those stylised facts, Allen et al. turn to a general equilibrium model and use their model to estimate the economic impacts of slavery. In that analysis, they find that:

In equivalent variation terms, slaveowners would have to pay an enslaved worker 1,192% more to get voluntary consent to the allocation of labor induced under slavery.

Breaking that down into three different sources of misallocation, they find that:

...allowing enslaved individuals to choose their own labor supply while remaining in slavery (i.e. removing coercion) yields a welfare gain equivalent to a 361% increase in wages; allowing enslaved workers to retain all of their earnings yields a welfare gain equivalent to a further 81% wage increase; and allowing enslaved workers to choose any occupation in any location in the U.S. (so that they act identically to free workers) leads to a welfare gain equivalent to a 55% wage increase.

Needless to say, those are massive welfare gains from removing slavery, but they likely understate the welfare costs of slavery. Importantly, their model does not capture all of the costs of slavery, including the physical and psychological harms associated with violence, abuse, and the loss of freedom. So, the welfare estimates from the model are best considered as lower-bound estimates of the full welfare costs of slavery (and the gains from removing slavery). It should also be noted that the percentage welfare gains from each source of misallocation compound rather than simply adding together, which explains why the overall welfare gain is so much greater than the sum of the three individual percentages.

Turning to the economy as a whole, Allen et al. then find that:

The movement of formerly enslaved workers out of agriculture causes services and manufacturing sectors to expand: for example, manufacturing output increases by 29% in the North, 19% in the South, and 27% overall. Agriculture expands in the U.S. North by 24%, but contracts in the U.S. South by 29%, for a modest decline of 5.4% overall. All told, aggregate GDP rises by 9.1%, with the reallocation of labor toward higher amenity locations offset by a combination of this structural transformation and the ability of the formerly enslaved to move to the (more productive) U.S. North.

A 9.1 percent increase in GDP is nothing to be sneezed at. Of course, this is based on the structure of the US economy in 1860, and can't be compared to what actually transpired later, after the Civil War.

Interestingly, but not unsurprisingly, not all workers would have been made better off by removing slavery. Free persons' welfare would decline by 0.7 percent overall, reflecting that many previously-enslaved workers would move to high-amenity and high-productivity areas, competing with free workers in those areas and lowering their welfare. However, that welfare loss pales in comparison to the welfare gains that previously-enslaved workers would receive, as well as the overall economic benefits.

US slavery was an immensely profitable institution for southern slave owners. There should be no doubt about that. The wealth accumulated through slavery remains visible today in the large plantation houses dating to the Antebellum period, some of which I may see after I cross the Mason-Dixon Line into the South for a couple of days next week. However, those plantation houses tell us little about the immense costs that slavery imposed on enslaved people, or the wider US economy. Allen et al.'s results provide some numbers to quantify those costs, even if they cannot capture the full human toll.

[HT: Marginal Revolution]