Job Advertising, Sourcing and Candidate Attraction: The Real Evidence on Referral Hires and Gendered Job Ad Wording

For a long time, where to advertise a vacancy was treated as an administrative choice rather than a question with real performance consequences attached. M. Susan Taylor and Donald Schmidt's 1983 study in Personnel Psychology treated it as exactly that kind of question. They tracked employees hired through seven different recruitment sources into a single manufacturing plant and measured how those sources differed once the person was actually on the job, on attendance, performance ratings and tenure, rather than on how many applicants each source produced or how cheaply. They found real, measurable differences between sources, and used the data to test two competing explanations researchers had proposed for why a recruitment source would matter at all after someone is hired.

One explanation holds that more effective sources simply transmit more complete, more accurate information about the job before anyone applies, so the person who accepts an offer already has realistic expectations and is less likely to be surprised, disappointed, or a poor match once the work actually starts. The other holds that sources do not work through information at all; they simply reach different populations of people to begin with, so a source that happens to attract more settled or socially embedded applicants will show better retention regardless of what it tells anyone. Taylor and Schmidt's own data, and the broader synthesis Michael Zottoli and John Wanous later published in Human Resource Management Review in 2000, tended to find more support for the information-based explanation than for the simple population-difference account, though neither fully displaced the other. That distinction is not academic hair-splitting: it points to two different levers a recruiter can actually pull. Improving what a job advertisement or a sourcing conversation communicates is within a recruiter's control. The underlying population a channel happens to reach is much less so, which is exactly the problem the rest of this lesson works through in more concrete terms.

Among the sources organizational researchers have studied most, the employee referral, a current employee recommending someone they know for an open role, has produced the most consistent record of better post-hire outcomes. That is not because referral is a clever hiring trick nobody else has thought to copy; it is because a referral changes what is happening at several points in the hiring process at once, and the research tradition behind it is old enough, and large enough, to say something specific about which point matters most.

Roberto Fernandez, Emilio Castilla and Paul Moore's 2000 study in the American Journal of Sociology, built on detailed cost data from a telephone call center, found that the firm's investment in its employees' own social networks produced measurable economic returns through lower screening, hiring and training costs. Referred candidates arrived effectively pre-screened by someone who already understood both the job and the person being recommended, work the organization would otherwise have had to do from scratch on every applicant.

Emilio Castilla's 2005 follow-up, also set in a call center and also published in the American Journal of Sociology, pushed the explanation further than a one-time screening advantage. Tracking referred and non-referred hires over time, Castilla found referred employees more productive immediately after hire, but the more consequential finding was that the advantage was not confined to the hiring moment at all. Referred employees who stayed socially connected to the colleague who referred them kept showing stronger attachment to the firm and sustained performance advantages well after any initial screening effect should have faded. The mechanism he describes is social rather than purely informational: a referred hire arrives with an existing relationship that keeps supplying support and a reason to stay, long after anyone has stopped checking whether the original referral was a good judge of talent.

Valery Yakubovich and Daniela Lup's 2006 study in Organization Science complicates the simpler claim that better referrers simply produce better hires. By breaking the hiring process into stages, they found that a referrer's own job performance predicts a referral's chances of passing some stages and not others. Where a decision rests on relatively objective, verifiable grounds, the referrer's performance is a meaningful signal; where the decision instead depends on a more subjective, relationship-driven judgment, the predictive link weakens. "Employee referrals work" is not one uniform mechanism; it is several distinct things happening at different points in a process that this research found worth pulling apart rather than treating as a single effect.

The most direct recent evidence on what referrals are actually worth to the hiring firm comes from Stephen Burks, Bo Cowgill, Mitchell Hoffman and Michael Housman's 2015 study in the Quarterly Journal of Economics, using personnel records from nine large employers across call centers, trucking and high-tech firms. Referred applicants were more likely to be hired and more likely to accept an offer once extended, even though referred and non-referred applicants looked similar on the skill characteristics the firms could measure directly, meaning the advantage was not mainly a matter of referrals being smarter or better-credentialed candidates on paper. Once hired, referred workers quit at substantially lower rates than non-referred workers across the firms studied, had fewer accidents in the trucking sample, produced more patents in the high-tech sample, and earned slightly higher wages. Put alongside Castilla's finding on sustained social attachment, the most defensible reading of this body of work is that referrals succeed mainly through better fit and information at the point of hire, and through ongoing social embeddedness afterward, not through privileged access to a pool of inherently superior candidates nobody else could have found.

None of this makes the employee referral an unambiguous recommendation for every organization in every circumstance, and a lesson reporting only the retention and performance findings above would give an incomplete, falsely reassuring account of the evidence. Roberto Fernandez and Isabel Fernandez-Mateo's 2006 study in the American Sociological Review examined exactly the mechanism that makes referrals valuable, an existing employee's personal network, and found that the relationship between network-based hiring and racial composition is not simply a story of excluded candidates lacking network access altogether. The more precise, and more troubling, finding is that when minority candidates did use personal networks to find work, those networks tended to perform worse for them than the same networks performed for other candidates.

Because a referral program is, structurally, a mechanism for an organization's current workforce to reproduce itself, an organization that is already demographically narrow and leans heavily on referrals for a large share of its hiring risks staying that way, or becoming more so, even while each individual referral decision looks sensible and even while the aggregate referral-outcome numbers reviewed above look favorable. A sourcing strategy that treats referral volume as an unqualified win, without tracking what it is doing to the composition of who even enters the applicant pool, is optimizing one outcome while quietly working against another the organization may also care about. This tension does not have a tidy resolution in the research; it is a real trade-off a hiring team has to monitor deliberately rather than one the referral evidence above makes disappear.

The research reviewed above supports a general, directional claim: referral hires tend to show a retention advantage over other common sources, for the mechanistic reasons already discussed. It does not license a precise, portable number that applies the same way across every organization, industry and labor market, and no single study gives a clean side-by-side ranking of referral against job boards, campus recruiting and staffing agencies on cost, speed and retention simultaneously. The figures below are illustrative only, constructed to teach the trade-off logic involved in choosing a sourcing mix, not drawn from any specific study's reported results.

Read this chart as a teaching device, not a benchmark to copy into a staffing report. Its shape, referral ahead of the other three, is consistent with the direction the peer-reviewed research above supports. The specific gap between bars, and the ranking among job boards, campus recruiting and agencies, is invented for illustration and will not match any particular organization's real data, which is exactly why a sourcing decision should be checked against an organization's own tracked outcomes rather than against a chart like this one, borrowed figures, or a competitor's claimed results.

Danielle Gaucher, James Friesen and Aaron Kay's 2011 study in the Journal of Personality and Social Psychology asked a question most organizations never examine directly: does the specific vocabulary a job advertisement uses, independent of the actual job, change who finds that job appealing? Across a series of studies, they first classified words commonly used in real job advertisements into masculine-coded terms, among them "competitive," "dominant" and "leader," and feminine-coded terms, among them "support," "understanding" and "interpersonal," then showed that job advertisements for male-dominated occupations already used more masculine-coded wording than advertisements for female-dominated occupations, independent of the job content itself. When they experimentally varied only the wording of an otherwise identical advertisement, participants rated the masculine-worded version as describing a more male-dominated occupation, and women, though not men, rated themselves as finding that job less appealing.

The mechanism Gaucher, Friesen and Kay identify matters as much as the headline finding. The effect did not run through women perceiving themselves as less skilled or less qualified for the role; it ran through a lower sense of belongingness, a perception that people like them do not fit in that kind of job. That distinction has a direct practical implication. A job advertisement is not only describing qualifications; its wording is also, whether an organization intends this or not, signaling who the role is implicitly written for, and that signal operates on a dimension, belonging, that has nothing to do with whether the reader could actually do the job well.

An honest treatment of this research also has to report what happened when other researchers tried to reproduce it in a different setting. Mihwa Seong's 2024 study in the Strategic Entrepreneurship Journal replicated the original design specifically in a start-up hiring context and found the effect held there, reducing young women's sense of belongingness much as the original study found. In an established-firm context, tested as a comparison within the same study, the same manipulation did not produce a statistically significant effect on either belongingness or perceived skill fit. That is a genuine boundary condition, not a failed replication to wave away: the underlying mechanism, wording shaping a felt sense of belonging rather than perceived competence, appears real, but how strongly it operates seems to depend on the kind of organization doing the hiring, a detail most advice built on the original 2011 study leaves out entirely.

An active candidate is someone currently searching: browsing job boards, submitting applications, telling contacts they are looking. A passive candidate is someone not searching at all, who might still move for the right opportunity if someone brought it to their attention in a way that felt worth their time. James Breaugh's 2008 review in Human Resource Management Review, surveying decades of recruitment research, treats the wording of the recruitment message, the timing of outreach, and the channel used as related but separable decisions, precisely because a message built to filter a large volume of self-selected applicants down to a shortlist is doing a different job than a message built to persuade one specific, currently uninterested person that a conversation is worth having.

This distinction has a direct practical consequence most job advertisements never account for. An advertisement optimized for active candidates can reasonably assume some baseline motivation already exists; its job is mostly to screen and inform. A message aimed at a passive candidate cannot assume that; it has to supply the motivation to pay attention at all, which is a persuasion problem, not an information-filtering problem, and a generic advertisement posted to a job board is simply the wrong tool for it. LinkedIn's "Why and How People Change Jobs" survey, published in 2015, reported that only a minority of professionals surveyed described themselves as actively looking for a new role at the time, while a much larger share said they would be open to hearing from a recruiter. That is an industry self-report survey, not a peer-reviewed study carrying the same methodological scrutiny as the research cited elsewhere in this lesson, and its specific figures should be read as directional industry data rather than an academic finding. It is nonetheless consistent, at the level of general pattern rather than exact number, with the basic distinction Breaugh's academic review and Zottoli and Wanous's (2000) recruitment-source synthesis both treat as real: most of the people an organization might want to hire are not, at any given moment, looking for a job at all.

The honest caveat worth stating plainly is that the specific, quantified claims circulating in recruiting practice about how much better passive candidates perform once hired rest on a thinner evidentiary base than the referral and wording findings reviewed earlier in this lesson, most of which come from peer-reviewed studies with firm-level personnel data. Treating the active-passive distinction itself as real and practically important is well supported; treating any particular industry-reported performance gap between the two groups as an established fact on the same footing as Burks et al.'s (2015) referral findings is not something this lesson's evidence supports doing.

Three limitations deserve to be stated directly rather than folded quietly into the sections above. First, most of the peer-reviewed sourcing-channel and referral research reviewed here comes from large, formalized employers in North American labor markets, call centers, a manufacturing plant, trucking firms, high-tech companies, with personnel systems detailed enough to support this kind of study. Whether the same referral advantage, and the same network-composition risk, holds with the same strength in smaller organizations, informal labor markets, or very different institutional and legal contexts for employment is not something this evidence was built to answer.

Second, this lesson does not resolve the tension between referral efficiency and workforce composition goals; it only states that the tension is real and needs active monitoring, not a confident formula for trading one against the other. An organization genuinely has to decide, case by case, how much weight to put on each, and nothing in Fernandez and Fernandez-Mateo's (2006) findings or Burks et al.'s (2015) retention numbers settles that question for them.

Third, the active-passive sourcing distinction is better supported as a real, practically important category than any specific quantified claim about how much better passive candidates perform once hired; the strongest numbers in general circulation on that second question come from industry surveys rather than from the kind of peer-reviewed, firm-level personnel data behind the referral findings earlier in this lesson, and the two kinds of evidence should not be treated with the same confidence.

Everything in this lesson concerns how a qualified person reaches, or fails to reach, an applicant pool: the channel the organization used, the words the advertisement chose, and whether the outreach matched an active search or had to create interest where none existed. None of it answers a different, equally consequential question: once a pool of candidates exists, sourced however it was sourced, which methods actually distinguish the candidates who will perform well from the candidates who will not, and how much confidence should an organization place in any single selection method's claimed accuracy. That is Lesson 38's subject, not this one's, and the two should not be collapsed into a single exercise even though a hiring team naturally experiences them as one continuous process. A well-sourced pool, built on the evidence reviewed here, still needs a selection process capable of actually telling its candidates apart; good sourcing cannot substitute for a weak or unvalidated assessment method, any more than a strong assessment method can fix a pool that never reasonably reflected the population an organization was trying to reach.

Select a real job advertisement you can examine closely: one currently posted by an organization you have access to, one you can obtain with permission from an HR contact, or one publicly posted that you can analyze responsibly without inventing information you do not actually have about the organization's sourcing mix.

Prepare a report of 2,000 to 2,500 words that:

1. identifies which recruitment sources the organization currently uses for this role (or, if that information is unavailable, states explicitly that it is unavailable rather than guessing), and evaluates the choice against the realistic-information and individual-difference explanations discussed in this lesson;
2. conducts a wording audit of the advertisement itself using the masculine-coded and feminine-coded categories from Gaucher, Friesen and Kay (2011), quoting the specific terms found and proposing specific alternative wording where relevant;
3. assesses whether the advertisement and its distribution channel are better suited to active candidates, passive candidates, or neither, and recommends a specific adjustment with a stated rationale;
4. discusses, using Fernandez and Fernandez-Mateo's (2006) findings, what risk, if any, the organization's current sourcing mix carries for workforce composition, and whether anything in the available evidence suggests that risk is being monitored; and
5. states explicitly where your own analysis required an assumption or was limited by unavailable data, rather than presenting an inference as a confirmed fact.

Use at least five academic sources, including at least three from this article's reference list. Treat any industry survey data you cite, including the LinkedIn survey discussed in this lesson, with the same caution this lesson applies to it: useful for directional context, not a substitute for peer-reviewed evidence.

Assessment rubric (100 marks)

1. Breaugh, J. A. (2008). Employee recruitment: Current knowledge and important areas for future research. Human Resource Management Review, 18(3), 103–118.
2. Burks, S. V., Cowgill, B., Hoffman, M., & Housman, M. (2015). The value of hiring through employee referrals. Quarterly Journal of Economics, 130(2), 805–839.
3. Castilla, E. J. (2005). Social networks and employee performance in a call center. American Journal of Sociology, 110(5), 1243–1283.
4. Fernandez, R. M., Castilla, E. J., & Moore, P. (2000). Social capital at work: Networks and employment at a phone center. American Journal of Sociology, 105(5), 1288–1356.
5. Fernandez, R. M., & Fernandez-Mateo, I. (2006). Networks, race, and hiring. American Sociological Review, 71(1), 42–71.
6. Gaucher, D., Friesen, J., & Kay, A. C. (2011). Evidence that gendered wording in job advertisements exists and sustains gender inequality. Journal of Personality and Social Psychology, 101(1), 109–128.
7. Seong, M. (2024). Does gendered wording in job advertisements deter women from joining start-ups? A replication and extension of Gaucher, Friesen, and Kay (2011). Strategic Entrepreneurship Journal, 18(2), 286–305.
8. Taylor, M. S., & Schmidt, D. W. (1983). A process-oriented investigation of recruitment source effectiveness. Personnel Psychology, 36(2), 343–354.
9. Yakubovich, V., & Lup, D. (2006). Stages of the recruitment process and the referrer's performance effect. Organization Science, 17(6), 710–723.
10. Zottoli, M. A., & Wanous, J. P. (2000). Recruitment source research: Current status and future directions. Human Resource Management Review, 10(4), 353–382.

Verification notes: every reference above was checked against publicly indexed bibliographic and abstract records (publisher and journal pages, University of Chicago Press, Oxford Academic/Quarterly Journal of Economics, SAGE, Wiley, INFORMS, and Google Scholar citation records) via web search rather than a full-text read of each paper. Findings are reported at the level of detail available in abstracts, publisher summaries and secondary accounts, cross-checked across at least two independent sources where possible. The 12-month retention chart earlier in this article is explicitly labelled illustrative and does not reproduce any single study's actual reported figures; no invented numeric comparison is presented elsewhere in this article as a real finding. LinkedIn's "Why and How People Change Jobs" (2015) survey is named in prose only, not in this numbered list, because it is an industry self-report survey rather than a peer-reviewed academic source, consistent with how earlier scripts in this series (for example script 33's handling of statutory citations) have kept non-academic factual material out of the academic reference list while still disclosing it plainly. The Choton Energy Solutions case, the comparison chart, the decision exercise, the myth/fact items, and the synthesis connecting this lesson to Lesson 36 (recruitment strategy and employer branding) and forward to Lesson 38 (selection methods and assessment validity) are original teaching material for this series.

Lesson 38 turns to the question this lesson deliberately leaves open: once a pool of candidates exists, sourced however it was sourced, which selection methods actually distinguish future strong performers from future weak ones, and how much confidence the evidence actually supports placing in any single method's claimed accuracy.