Why the Census Job Could Outlive the Census
The 2010 Decennial Census hiring program is the key case for understanding how a temporary public job can become more than a short-term paycheck. This chapter follows the route from application to verified Census work, then asks why the score threshold used in the study can reveal a local employment effect without making a score of 70 a job guarantee. The practical value is a clearer view of which parts of a public hiring process might leave a credible trace for the worker who comes next.
The Census Bureau recruited at a scale that made the program unusual even among public employment efforts. Between June 2008 and May 2010, it recruited 3.9 million applicants for more than 850,000 temporary positions. The jobs included enumerators, office clerks, crew leaders, supervisors, recruiting assistants, and related roles. The positions supported a mandatory national count whose results determine congressional representation and help allocate federal funds. That public purpose matters: the work was not an arbitrary simulation of employment, even though most positions were temporary.
A national count built a hiring credential
For the non-supervisory positions studied in the paper, the process began with an application and an employment test. The test took 30 minutes, with additional time for paperwork. Applicants generally needed a converted score of 70 to be considered for employment. The threshold did not measure every attribute required on the job, and it did not finish the selection process. It was an early gate in a longer chain.
After testing, applicants underwent a background name check. Those who passed were placed on a selection sheet when they matched a particular opening. An opening was defined not just by a job title but by the position, location, days, hours, and transportation required. This detail is important because availability was part of the treatment opportunity. A person who could not reach a workplace or work its schedule was not equivalent to someone who could.
Selection sheets ranked eligible applicants by converted test score and veteran status. Veterans received a preference, so the research excludes them from the comparisons. Among applicants with the same score and veteran status, the ordering was random. Hiring managers were instructed to interview from the top downward, separately for new and experienced applicants, until they could make enough offers. The manager still assessed interest, availability, and qualification. A high score increased a person's position in the queue; it did not force the manager to make an offer.
The process continued after an offer. Applicants completed training, were fingerprinted, underwent another background check, and had their employment eligibility verified. In the study's data, a person counts as hired when the researchers observe Census hours and pay. This operational definition keeps an offer, a cleared applicant, and an actual worker distinct. Each status could carry a different meaning for a future employer or for the worker's own sense of attachment to formal work.
That sequence creates several possible channels for a later effect. Training can add skills. The job itself can provide a recent work history after a period of unemployment. A manager or colleague might offer a reference. The position can expose a worker to the routines of a formal workplace. And the combination of a test, background checks, and a trusted public assignment can make the fact of being hired legible as evidence that the worker met a standard beyond simply accepting any available activity.
The last possibility is the most distinctive. The Census authors consider whether the long-run employment gain for marginal applicants came from experience and training or from the signal of passing screening and carrying out work that requires trust. The evidence does not measure what a later employer believed when seeing a Census job on a résumé. It only makes the signal hypothesis plausible. That distinction matters: a plausible channel is not the same as a demonstrated mediator.
The national character of the work may also have strengthened its meaning. A Census assignment has a clear public function and a recognizable institution behind it. The study does not show that every employer consciously rewards a Census credential, nor that every temporary government job has the same reputation. It shows why the institutional details belong in the explanation. “Temporary public employment” describes the duration and sponsor, but not the information the job conveys.
The sequence has two stopping points that are easy to miss. First, the score affects the probability of reaching the later stages; it does not determine the entire path. Second, the study observes the later employment outcome but cannot watch a specific employer interpret the Census experience. The diagram therefore ends with an unresolved signal or experience channel, not with a claim that the hiring process automatically caused every later job.
What the threshold can and cannot tell us
A threshold is useful for evaluation because people very close to it can be similar while facing different chances of entry. Imagine two applicants whose scores are just on opposite sides of 70. If the cutoff raises the chance that one receives a Census job, and if nothing else changes abruptly at that exact point, the later difference in employment can be linked to that additional hiring opportunity more credibly than a simple comparison of all hires with all non-hires.
This is the intuition behind a fuzzy regression discontinuity. “Regression discontinuity” refers to the comparison near a cutoff. “Fuzzy” means the cutoff changes the probability of treatment rather than assigning treatment perfectly. The Census process supplies several reasons for that fuzziness: the threshold was generally required for consideration, but interviews assessed availability and qualification; applicants could appear on different selection sheets; and an offer was not the same as observed hours and pay.
The first stage—the effect of crossing the cutoff on actual hiring—was about 6 percentage points in the study's local sample. That may sound small compared with the employment effect reported later, but it is enough to create a strong instrument when the threshold is otherwise credible. The researchers use that change in hiring to estimate the effect of Census employment for the applicants whose hiring status was shifted by being just above rather than just below the cutoff.
The “for whom” is the critical phrase. The estimate is local to applicants near the score-70 margin, not an average for everyone who applied. Those applicants had lower prior employment, lower prior W-2 income, and far lower college completion than high-scoring applicants. The score-70 design therefore reaches people whose outside options may have been weak. It tells us what the temporary Census opportunity did for that margin, not what it would do for an experienced worker who already had several strong offers.
The study performs several checks on that comparison. Before Census hiring began in 2009, applicants just above and below the cutoff had similar labor-market outcomes. Observable characteristics such as college completion and selective-service registration do not show meaningful jumps at the threshold, and the estimated first stage is very strong. The researchers also inspect retaking behavior. If applicants had known precisely how to manipulate their scores to cross the line, one might expect a jump in test retakes just above 70. Instead, the estimated change in retaking is small and negative, with a confidence interval that does not include an increase.
Not every check is equally conclusive. A formal test of whether applications bunch at the threshold produces results whose direction differs between conventional and robust versions. The authors argue that the test fits poorly because application counts grow rapidly at low scores and then follow a more linear pattern. They rely more heavily on the visual distribution, retake evidence, balance tests, and the absence of pre-existing outcome differences. This is a reasonable basis for confidence in the design, but not a license to forget that the method depends on assumptions about what changes smoothly at the cutoff.
There is also a measurement caution around demographic data. Some 2010 Census responses could have been affected by applicants' experiences of being interviewed or hired for the Census. The study finds changes in nonresponse at the threshold and therefore interprets those demographic comparisons carefully. That caution does not erase the employment result, which is based on linked administrative records, but it shows why a credible design in one outcome does not automatically validate every variable measured around the same cutoff.
The threshold does not answer the same question as the selection-sheet comparison used for higher-scoring applicants. In the broader design, people with the same score were available for the same kind of opening, and their random position among equal-score applicants affected the chance of an initial offer. That approach helps study applicants beyond the bottom of the test distribution, but it describes another local population. The threshold result and the within-score result should be placed next to each other, not blended into one universal Census effect.
What, then, might travel from this case to another public job? A transparent standard can make access less dependent on opaque discretion. A formal screening and verification process can create a credential that employers understand. A real public task can provide work that is easier to describe and verify than a make-work activity. A job-specific selection process can match the opportunity to schedule and location. These are design possibilities suggested by the case, not findings that the study separately estimates.
What is less portable is equally important. The Census was a trusted national function during a weak labor market. It hired many people with relatively high skill levels alongside applicants near the lower end of the score distribution. The jobs required testing, interviews, background checks, and eligibility verification, and the paper describes the pay as relatively good for the required skill level. A temporary program without those features may not send the same signal. A job that has no recognizable employer, no meaningful task, or no route to a subsequent workplace could produce a different trace even if its duration were identical.
The threshold is therefore best understood as a doorway into a specific institutional sequence. It helps the study identify what happened when marginal applicants gained a higher chance of entering that sequence. It does not tell us that the number 70 itself created the long-run effect, or that crossing any administrative line will do so. The lasting question is what became possible after entry: was the worker more employable because they had acquired experience, because they could credibly report a screened public role, because someone could vouch for them, or because several of these changes reinforced one another?
The resulting employment evidence is striking precisely because it is uneven. The strongest effect appears near the margin where a public job may replace no job at all. For applicants with stronger alternatives, the same opportunity mostly creates a short-term interruption rather than a persistent advantage. That contrast turns the hiring process into more than background: it explains who could enter, what the entry meant, and why the result cannot be read as a general effect of temporary work.
The central lesson is modest but consequential. A public job can outlive the public project when the process around it gives the worker a credible and usable position in the labor market. The study can identify the employment consequence of crossing the threshold for marginal applicants; it cannot settle whether the durable channel was signal, experience, confidence, or some combination. The next comparison must therefore follow the outcome gradient itself: why did the job matter much more to applicants with the weakest alternatives than to people who were likely to work anyway?