What the Evidence Can Actually Identify
The Austrian study is asking a causal question: what happens to a vacancy when applicant supply changes, holding apart the features that made the vacancy attractive in the first place? That distinction matters because a large application count can be a symptom of a desirable job rather than the cause of a better hiring outcome. The evidence is useful only if we can tell those explanations apart. This chapter follows the study’s mass-layoff design and shows what its estimates identify, what they leave as an assumption, and why the results should be read as local rather than universal.
Why application counts are not self-explanatory
Suppose two vacancies receive different numbers of applications. It would be easy to compare their filling times or the tenure of their hires and attribute the difference to the application count. But vacancies do not attract applicants at random. A well-known firm may draw more interest. A job with higher pay, better hours, or a more attractive location may both receive more applications and fill more easily. An occupation facing a temporary shortage may affect applications, wages, and duration at the same time. Seasonal hiring could create similar patterns through the timing of both job postings and layoffs.
These are confounding influences: factors related to both the number of applications and the outcome being measured. Firm practices can matter as well. An employer that writes clearer ads, responds quickly, or offers a more realistic job description may attract more applicants and also create better matches. A simple comparison would not reveal whether the applications changed the firm’s options or merely accompanied other advantages.
The paper addresses this problem by looking for an outside change in applicant supply. It uses the inflow of registered jobseekers from mass layoffs in the same four-digit sector and district during the three months before a vacancy is posted. The baseline measure counts jobseekers from firms that lost at least 20% of their workforce in a single month. A sudden local layoff can increase the number of people available to apply to nearby vacancies, even when the vacancy itself did not change its pay, requirements, or employer.
The design is not claiming that layoffs are harmless events. A mass layoff can affect a local economy in many ways. The identifying idea is narrower: after accounting for the study’s controls, the relevant pre-posting inflow changes the hiring outcomes through the supply of applicants rather than through some other route. That is an assumption to evaluate, not a fact guaranteed by the word “shock.”
The geographic and sectoral detail is part of the logic. A national wave of unemployment might coincide with broad changes in demand, wages, or business conditions. A sector-and-district inflow is intended to capture a more targeted change in the pool facing a vacancy. The study also sums the inflow over the three months before posting, connecting the shock to the period when the vacancy’s search begins rather than to an indefinite background condition.
The regressions add firm, occupation, and posting-month fixed effects. In plain terms, the comparison is made after absorbing persistent differences in how firms hire, differences between detailed occupations, and common timing patterns. The analysis also controls for the days between posting and the expected job start. These adjustments do not make the design experimental, but they reduce several predictable ways in which vacancies with different application counts might differ.
From a local shock to a causal estimate
The layoff measure must first change the observed number of applications. This is the first stage of the design. It does: the reported first-stage coefficient is 0.275, with a standard error of 0.007, and the associated F-statistic is about 1,400. In the paper’s inverse-hyperbolic-sine specification, doubling the local jobseeker inflow raises first-week observed applications by approximately 0.19, or roughly 3–4% of the mean of 5.6 applications.
The importance of a strong first stage is straightforward. If the proposed supply shock barely changed applications, it could not tell us much about the effect of applicant supply. Here the relationship is strong and precisely estimated. But relevance is not the same as validity. A good first stage shows that mass layoffs move the proposed treatment; it does not by itself show that they affect vacancy outcomes only through that treatment.
It helps to separate three comparisons that are easy to collapse. The raw comparison asks whether vacancies with more applications have different outcomes. The reduced-form comparison asks whether vacancies exposed to a larger mass-layoff inflow have different outcomes. The IV estimate combines the latter change in outcomes with the first-stage change in applications to estimate the effect of the applicant-supply movement induced by the shock. That estimate is closer to a counterfactual question—what would have happened to an affected vacancy with a smaller versus a larger pool—but only for the vacancies and marginal applicants moved by this particular source of variation.
This interpretation also explains why the paper reports effects per observed application and per standard deviation rather than presenting the coefficient as a universal return to recruiting effort. An employer cannot assume that adding an applicant through a private platform produces the same change as the marginal applicant supplied by a nearby mass layoff. The source of the variation matters because it selects which employers, locations, occupations, and workers are responsive. The design is valuable precisely because it identifies a concrete margin, not because it removes all context from the estimate.
The study checks whether the mass-layoff measure is already related to characteristics that should have been determined before the vacancy’s applicant pool changed. Firm workforce composition, firm size, working time, job type, required education, and the expected start date are among the characteristics examined. Most standardized relationships are economically very small—generally no more than about half a percent of the characteristic’s standard deviation for a one-standard-deviation change in the instrument. That pattern is reassuring because it suggests the shock is not simply sorting into visibly different vacancies.
There is an important qualification. With more than a million vacancies, very small differences can be estimated precisely. The paper’s joint balance test rejects exact balance, even though the individual standardized effects are mostly tiny. That result does not overturn the design, but it is a reason to describe the balance evidence accurately: the instrument is broadly similar across observed predetermined characteristics, not perfectly independent of every one of them.
The causal interpretation also depends on an exclusion restriction. The study needs mass layoffs to affect filling time, tenure, wages, or worker composition through applicant supply, conditional on its controls and fixed effects. If layoffs independently changed the firm’s demand for labor, the local cost of operating, or the quality of the jobs being posted, the IV estimate could include those channels. The data cannot prove that no such pathway exists. The design makes the assumption more plausible through timing, controls, and robustness checks; it does not turn the assumption into an observation.
A second measurement assumption concerns what is actually counted. The applications in the study come from referrals made by the public employment service during a vacancy’s first week. They are not a complete administrative record of every application submitted through every channel. The paper treats them as a proxy for total applicant supply. Under the assumption that referred and outside applications are equally likely to fill a vacancy, about 18% of filled vacancies being filled by a referred jobseeker implies an estimated first-week total pool of roughly 31 applications per vacancy.
That scaling is useful for intuition but should not be mistaken for a direct count. It changes the unit in which a per-application coefficient is expressed. It does not change the standard-deviation effects because both the coefficient and the application variation scale together. More importantly, the proxy must be representative enough for a change in referred applications to stand in for a proportional change in the unobserved total pool. If referrals are systematically different from outside applications, the interpretation becomes less secure.
The design therefore identifies a particular causal margin. It estimates how outcomes respond for vacancies whose applicant pools increase when nearby mass layoffs send jobseekers into the market, and for the marginal applicants supplied through that channel. This is often called a local effect. It is not the effect of forcing every employer to receive one additional applicant, nor the effect for every kind of worker or vacancy.
The chain in the diagram has two different kinds of links. The solid links describe the intended source of variation: layoffs increase the local flow of jobseekers, which changes observed applications, which is then related to outcomes. The dashed links are assumptions that make the interpretation legitimate. Controls can absorb measured differences, but they cannot observe the exclusion restriction. A proxy can move reliably, but it still needs to represent the broader quantity the researcher wants to study.
Robustness narrows the alternatives
The paper tests whether the instrument is capturing a persistent local condition rather than a temporary applicant-supply change around the time of posting. If a district were simply a difficult place to hire in for many months, layoffs occurring long after the vacancy was posted might also predict its outcomes. Instead, layoffs in the pre-posting window affect the outcomes, while layoffs seven to nine and ten to twelve months after posting—after the typical vacancy is filled—do not significantly affect wages or vacancy duration. The earlier later-window placebo has at most a marginal effect on tenure.
This timing pattern supports the idea that the relevant event is a surge of applicants close to the start of the search. It does not prove that the pre-posting layoffs have no other local effects, but it weakens an explanation based only on a durable background feature of the district or sector. A shock that matters when applicants can reach the vacancy, and not after the hiring decision has passed, fits the proposed channel better than a general label for local labor-market difficulty.
The paper also changes the threshold used to define a mass layoff, from a loss of at least 10% of a firm’s workforce to thresholds as high as 50%. The first stage is stronger at lower thresholds, while the estimated effects keep the same sign across the range and remain significant at all but the most extreme cutoffs. The magnitudes vary, and larger shocks can identify somewhat different marginal applicants. That variation is not noise to hide; it reminds us that an IV estimate describes the people and vacancies moved by the particular shock used.
Taken together, the checks make the applicant-supply interpretation more credible than a raw correlation. They do not answer every external-validity question. The sample comes from Austrian vacancies on a public employment-service board, and the linked worker records cover a subset of postings. The referral measure may not represent applicants who use private platforms, personal networks, or direct employer contact. The paper’s own local design tells us to transfer the mechanism more confidently than the exact coefficient.
This is the right standard for reading the results. A strong first stage supports the claim that the instrument changes applications. Small balance differences and placebo behavior help assess whether it is acting through the intended timing and pool. The exclusion and representativeness assumptions remain the bridge from those facts to a causal interpretation. Once that bridge is acknowledged, the next question can be asked with appropriate precision: when applicant supply rises for this identified group of vacancies, does the study see faster filling, longer-lasting employment, or both?