What Spence Actually Told Us
Michael Spence published a paper in 1973 that changed how economists think about hiring, education, and the whole credentialing apparatus that runs through modern labor markets. The core idea is simple enough that it feels almost trivial until you watch it play out in real life. Workers send signals to employers about their productivity, and those signals take the form of degrees, certifications, test scores, and other credentials that are costly to obtain. The cost difference is what matters. High-productivity workers can acquire credentials at a lower personal cost than low-productivity workers, which makes the credential a credible signal rather than just noise. I spent about seven years working in technical recruiting before moving into people analytics. The first thing you notice when you're actually on the hiring side is how much weight gets placed on signals that have almost nothing to do with the job itself. A degree from a particular university, a certification from a professional body, a specific GPA threshold. These things act as screening mechanisms because employers cannot directly observe productivity before the hire. The credential is a proxy, and proxies are imperfect by definition. The mechanism works through what Spence called the separating equilibrium. In a separating equilibrium, different types of workers choose different signal levels, and the employer can read the signal to infer the underlying type. High-productivity workers get the credential. Low-productivity workers do not, because the cost of obtaining it is too high relative to their expected benefit. This is the whole game. But the model has assumptions that do not hold up well in practice, and that is where things get interesting.
The cost of signaling is not just monetary. It is time, effort, opportunity cost, psychological strain. A student taking an extra year to complete a degree is giving up a year of earnings, plus the mental bandwidth that could have gone toward actual work experience. Employers know this implicitly. They build credential requirements into job descriptions not because the credential teaches job-specific skills, but because it filters for traits like persistence, conformity, and the ability to navigate bureaucratic systems. Those traits are real and valuable, just not in the way a syllabus would suggest.
Where the Model Breaks Down
One thing beginners miss about Spence's framework is that it assumes employers can perfectly observe the signal and perfectly interpret it. In reality, employers observe noisy signals and interpret them through institutional biases, market conditions, and their own limited information. A 3.7 GPA from a community college does not carry the same weight as a 3.7 GPA from an Ivy League school, even if the underlying productivity is identical. The signal is contaminated by the sender's background, which is not what the model predicts. Another issue is credential inflation. As more workers obtain credentials, the signaling value of those credentials declines. Employers respond by raising the threshold. What was once a bachelor's degree requirement becomes a master's degree requirement. What was a certification becomes two certifications. This is the arms race that Spence's model implies but does not fully explore. The result is a labor market where credentials function less as signals of productivity and more as barriers to entry that benefit incumbent credential holders. I ran into this directly when I was redesigning our company's engineering hiring process around 2019. We had been requiring a computer science degree for all backend roles. The pipeline was strong but homogenous. About 80 percent of applicants came from traditional four-year programs. We started looking at portfolio-based evaluation as an alternative signal. GitHub repositories, open-source contributions, technical blog posts, completed projects. These are messier signals than a degree. They are harder to standardize across candidates. But they correlated better with actual job performance in our specific context, which is what matters.
Get the Full Details
The workaround I implemented was a hybrid screening model. Degree requirements stayed for the initial resume screen to maintain throughput. But we added a portfolio review tier that allowed non-traditional candidates to bypass the degree filter if their demonstrated work met a defined quality standard. We created rubrics for code quality, project scope, documentation, and collaboration history. This took about three months to design and validate. Once it was in place, we saw a 40 percent increase in candidate diversity without any drop in hiring quality metrics over a 12-month period.
The Pooling Equilibrium Problem
Spence also described the pooling equilibrium, where all worker types choose the same signal level and the credential loses all discriminatory power. This happens when the cost of signaling is roughly equal across productivity types, or when the return to signaling is so high that even low-productivity workers find it worthwhile to obtain the credential. The modern labor market has moved significantly toward pooling in several sectors. A bachelor's degree is now so common that it barely functions as a signal at all. It is a floor, not a differentiator. When pooling occurs, employers shift to secondary signals. Internships, referral networks, brand-name employers on your resume, specific technical stacks you have worked with. These are thinner signals, meaning they carry more information per unit, but they also reproduce existing inequalities. Referrals favor people who already have professional networks. Brand-name employers favor people who got their first break at the right company. The credential system does not eliminate bias. It just moves it to a different layer. There is a counter-intuitive insight here that most people miss. The signaling model suggests that more information leads to better outcomes. But in labor markets, more signaling often leads to worse outcomes because the signal itself becomes the product rather than a reflection of underlying ability. Bootcamps, certification mills, and degree mills exist precisely because there is demand for the signal independent of the skill. Employers are aware of this to varying degrees, but they continue to rely on credentials because the alternative, actual skills assessment, is more expensive and less scalable in the short term.
The tradeoff is real. Skills-based screening takes more time per candidate. It requires more trained evaluators. It introduces more subjectivity into the process, which creates legal and compliance risks for large organizations. A degree is a standardized, legally defensible, easily comparable signal. It is also a poor predictor of job performance in many cases. The fact that organizations continue to use it says more about institutional inertia than about rational optimization.
How to Work With the System
If you are a worker trying to navigate a signaling-dominant labor market, the practical move is to understand which signals your target employers actually value and which are theater. Research the hiring patterns of companies you are interested in. Look at the backgrounds of people currently in the roles you want. Identify the common credentials and evaluate whether they correlate with actual competency or just with access. If the correlation is weak, invest in signals that have higher information content for your specific field. For technical roles, a strong public portfolio often outperforms an advanced degree from an unknown program. For creative roles, published work and demonstrated client relationships outweigh certificates. For regulatory fields, licenses are non-negotiable and the signaling value is real because the cost of failure is so high. The key is matching the signal to the signal environment of your target market. Blindly accumulating credentials without understanding what employers actually use to screen is a common mistake that wastes time and money. If you are an employer relying on credential-based screening, the recommendation is to periodically audit your requirements against actual performance data. Track the correlation between credential indicators and on-the-job outcomes. If the correlation is weak, the requirement is likely just filtering for privilege rather than productivity. Adjust your screening criteria accordingly. This is not easy to do because it requires data infrastructure and a willingness to challenge established hiring practices. Most organizations skip this step because the cost of doing it is higher than the cost of continuing with suboptimal screening.
The Limits of the Framework
Spence's model is a foundational piece of labor economics, but it should not be treated as a complete theory of the labor market. It does not account for network effects, social capital, discrimination, geographic constraints, or the role of luck. Two workers with identical credentials and identical productivity can have very different career outcomes based on factors the model does not capture. The signaling framework is useful for understanding one mechanism among many. It is not a general theory of employment. The model also assumes rational actors with perfect information about signal costs and benefits. Real workers make credential decisions based on incomplete information, social pressure, family expectations, and limited financial resources. A first-generation college student does not face the same signaling calculus as someone with parents who navigated higher education successfully. The model's assumption of rationality masks these structural differences. Despite these limitations, the framework remains useful for thinking about why credentials matter, how they function as filters, and what happens when the signal becomes decoupled from the underlying trait it is supposed to represent. The labor market will continue to rely on signals because direct assessment of productivity is expensive and difficult. The question is not whether signals exist but whether the signals being used are actually informative or merely conventional. That distinction determines whether the system allocates talent efficiently or simply reinforces existing hierarchies.
