Knowledge that weren’t collected prior to now are sometimes unavailable to researchers and policymakers within the current. A current and fast-growing literature is trying to get round this limitation by amassing info buried in outdated newspapers. Digitised newspaper articles from the previous comprise info on wages, costs, and occasions comparable to strikes, episodes of violence, and pure disasters. These articles supply fine-grained geographic and time variation as nicely. Thus far, historic newspapers have been used to extract info on, amongst different subjects, anti-Black sentiment (Ottinger and Winkler 2022), fertility restrictions (Seashore and Hanlon 2019), costs and varieties of cotton seeds (Rhode 2021). As with information extracted from maps by way of machine-learning strategies (Combes et al. 2020), the rising availability and scope of digital newspaper archives (e.g. newspapers.com or Chronicling America) opens avenues to generate long-run information beforehand unavailable to researchers. This growth follows a current pattern of integrating historic information and strategies into mainstream economics (Margo 2017).
Sometimes, researchers gather newspaper-based information for the aim of utilizing them as final result, remedy, or management variables in statistical evaluation. In a current paper (Ferrara et al. 2022), we present how information generated from historic newspaper articles can be utilized for one more vital objective: to resolve measurement error in statistical evaluation. We construct on the framework by Chalfin and McCrary (2018), who argue that when a researcher has two mis-measured variables for the same amount of curiosity, one measure can function an instrument for the opposite and recuperate the true coefficient of curiosity so long as the errors within the two variables are uncorrelated. Nonetheless, that is not often an possibility, as amassing a second impartial measure for the same amount of curiosity tends to be prohibitively costly. We present {that a} second such measure could be cheaply generated from digitised newspapers, and description the circumstances beneath which such a method is probably going to reach sensible and empirical phrases.
As an instance our framework, we replicate two current research on the impact of the unfold of the boll weevil on financial outcomes within the US South between 1892 and 1922 by Ager et al. (2017) and Clay et al. (2019). The unfold of the boll weevil is usually measured from a map that was revealed by the US Division of Agriculture (USDA). Whereas this map is usually of top quality, it accommodates errors such because the crossing of date traces proven in Determine 1. Every of those traces marks the furthest unfold of the cotton-boll consuming beetle in any given 12 months; theoretically, the date traces can not cross. If such errors happen at random, any form of statistical evaluation would underestimate the affect of the boll weevil. An correct estimate of the pest’s impact is vital as a result of it informs policymakers who want to use it as a baseline comparability for the unfold of bugs in different contexts right now.
Determine 1 Errors within the USDA map for the unfold of the boll weevil
Notes: The USDA map for the unfold of the boll weevil reveals the arrival date of the beetle in annually throughout Southern counties between 1892 and 1922. Every line marks a distinct arrival 12 months, which means in precept that these traces shouldn’t cross. In follow, such crossings happen; examples are marked by the purple rectangles.
As a way to resolve this measurement error drawback within the Chalfin and McCrary (2018) framework, one would require a second measure of the boll weevil unfold over time and house that may very well be collected at an affordable value. We generate such a measure from digitised newspapers by looking articles in newspapers.com that embody the phrases “boll weevil” and the identify of the county for which we search to measure the pest’s arrival. Whereas not all counties have recorded newspaper articles, newspapers in the identical state would report appearances of the weevil in different components of the state as within the examples offered in Determine 2.
Determine 2 Boll weevil reporting in native newspapers
Notes: Newspaper stories of the arrival of the boll weevil in Marion County, Mississippi, by the Jackson Each day Information in Hinds County (left) and the Star Ledger in Attala County (proper).
We assemble a loud newspaper-based measure of the boll weevil’s arrival date based mostly on the utmost of the five-year transferring common within the share of papers mentioning the pest along with a county’s identify. We offer an instance of the newspaper-based arrival measure compared to the USDA map date for Marion County in Determine 3. The newspaper information are typically noisy, which is why we easy out a number of the noise by making use of the five-year transferring common. What needs to be famous is that for our framework to carry, it isn’t vital whether or not the newspaper information present a roughly noisy measure of the boll weevil’s arrival date than the USDA map. The one required assumption is that errors within the map should be uncorrelated with errors within the newspaper protection of the vermin.
Determine 3 Instance of the newspaper-based and USDA arrival dates for Marion County, MS
Notes: The dashed line is the share of articles mentioning “boll weevil” and “Marion County” amongst all articles mentioning “Marion County” in each obtainable newspaper outlet in Mississippi between 1882 and 1932. The strong line represents five-year transferring averages of this share. The purple horizontal line reveals the boll weevil’s arrival in Marion County from the USDA map. The blue horizontal line signifies the expected arrival on the most of the five-year transferring common.
We define 3 ways by which this second measure can be utilized to deal with measurement error within the USDA map arrival date variable. This contains set identification, pattern restrictions, and a parametric bias correction. Probably the most intuitive of those is the pattern restriction, the place we use observations for which the arrival date within the USDA map and within the newspaper-based measure coincide. We name this the ‘settlement pattern’. Whereas there’s a likelihood that both of the 2 measures is improper, the prospect that each are collectively improper has a considerably decrease chance.
The primary takeaways from our replications of Ager et al. (2017) and Clay et al. (2019) are the next. First, even when utilizing the newspaper-based arrival date as a substitute of the USDA date, we are able to nonetheless replicate the findings in each papers. Had the USDA map not existed, their research might have been carried out with info gathered from digitised newspapers. This highlights the usefulness of digitised historic newspapers for the creation of novel information content material. Second, utilizing the newspaper-based arrival date by making use of our three measurement error correction strategies elevated the impact sizes and strengthened the leads to each papers. Regardless that the newspaper-based measure is coarse and was generated in a quick and inexpensive approach, it offered substantial worth in decreasing measurement error within the unique USDA map.
Oftentimes, researchers are likely to ignore measurement error in utilized settings so long as some typical stage of statistical significance is achieved. When this isn’t the case, probably promising analysis initiatives are typically deserted. We hope to offer a low-cost various for coping with measurement error, particularly in settings that use historic information the place measurement error is a pervasive drawback. Our technique utilizing newspaper-based info works finest for portions that may be simply extracted from newspapers, comparable to occasions that may be readily recognized with single (or a small set of) search phrases. Different portions, comparable to costs, are considerably more durable to extract; our options will not be a possible avenue in these instances.
References
Ager, P and B Herz (2019), “From the farm to the manufacturing facility ground: How the structural transformation triggered the fertility transition”, VoxEU.org, 16 Might.
Ager, P, M Brueckner and B Herz (2017), “The boll weevil plague and its impact on the southern agricultural sector, 1889–1929”, Explorations in Financial Historical past 65: 94–105.
Seashore, B and W Hanlon (2019), “Censorship, household planning, and the historic fertility transition,” VoxEU.org, 4 August.
Chalfin, A and J McCrary (2018), “Are U.S. Cities Underpoliced? Principle and Proof”, Overview of Economics and Statistics 100(1): 167–186.
Clay, Okay, E Schmick and W Troesken (2019), “The Rise and Fall of Pellagra within the American South”, Journal of Financial Historical past 79(1): 32–62.
Combes, P-P, G Duranton, L Gobbilon, C Gorin and Y Zylberberg (2020), “(Resolution) timber and (random) forests: City economics, historic information, and machine studying”, VoxEU.org, 17 November.
Ferrara, A, J Y Ha and R Walsh (2022), “Utilizing Digitized Newspapers to Refine Historic Measures: The Case of the Boll Weevil”, NBER Working Paper No. 29808, February.
Margo, R (2017), “The combination of financial historical past into economics”, VoxEU.org, 3 September.
Ottinger, S and M Winkler (2022), “The Political Economic system of Propaganda: Proof from US Newspapers”, IZA Working Paper No. 15078, February.
Rhode, P (2021), “Organic Innovation with out Mental Property Rights: Cottonseed Markets within the Antebellum American South”, Journal of Financial Historical past 81(1): 198–238.
