The Same Switch, Different Gain
The central empirical question is whether changing employers is enough to show that workers are climbing the same ladder. The evidence says no. The Chicago Fed study finds nearly identical job-to-job transition rates for men and women, but different employment continuity, different destinations, and markedly different pay growth once the starting wage is taken into account. Reading these measures together shows what a mobility statistic can establish—and what it leaves out.
A headline rate leaves the destination blank
A job-to-job transition is a direct move from one employer to another. It is a meaningful event: the worker has found, or accepted, an outside opportunity without first becoming nonemployed. But the rate is a narrow measure. In the paper's transition figure, the job-to-job rate is the percentage of people employed in the prior month who move directly to another job. It does not describe the pay of the new job, the hours attached to it, the amenities it provides, or the worker's position before the move.
Using the Current Population Survey, the Survey of Consumer Expectations, and the Survey of Income and Program Participation, the researchers find no statistically significant gender difference in this rate. The result persists when they control for observable demographics and occupation. “No statistically significant difference” does not mean that every estimate is exactly equal or that every occupation offers the same opportunities. It means the data and the specified comparisons do not establish a reliable average difference in the measured direct-transition rate.
That narrow result is easy to overread. If women and men appear equally likely to move directly from one employer to another, it is tempting to infer that they face equal opportunities to advance. But advancement is not what the rate measures. Two people can each make one direct move while landing in jobs with different hours, pay, productivity, flexibility, or room for later raises. A count of moves treats those outcomes as one event.
The surrounding employment flows reveal why the distinction matters. An employment-to-nonemployment transition, often called an E-to-N flow, takes a person out of a job. A nonemployment-to-employment transition, or N-to-E flow, brings someone back into work. The paper finds that women are more likely to leave employment and less likely to re-enter than men. In its CPS sample, the employment-to-population ratio is 79.1% for men and 66.5% for women.
These flows use different denominators from the job-to-job rate. An E-to-E rate asks what share of people who were employed moved to another job. An E-to-N rate asks what share of employed people became nonemployed. An N-to-E rate asks what share of nonemployed people found work. Putting them beside one another prevents a common error: interpreting an equal rate among people already employed as if it described the entire path through the labor market.
Imagine, only as a measurement example, that ten out of every hundred employed workers in each group switch employers. The E-to-E rates match. That tells us nothing about how many people in each group were employed before the comparison began, how many left work afterward, or how many nonemployed people could re-enter. Nor does it tell us whether the ten moves went to jobs with the same hours or pay. The rate is conditional on being employed; the broader ladder is shaped by who remains in that denominator and what happens after the move.
A similar caution applies to the phrase “opportunity to advance.” A direct move can be voluntary, prompted by an outside offer, or part of a search for a better match, but the transition measure does not rank those motives or outcomes. It also does not establish that a worker who stays has failed to progress: they may receive a raise or improve their position without changing employers. Conversely, a worker who moves may exchange pay for flexibility or may enter a less stable job. Counting transitions without observing the alternative paths makes all of these cases look alike.
Employment continuity affects the ladder in at least two ways. First, a period without employment can reduce current earnings even if the next job is obtained directly or relatively quickly. Second, re-entry can place a worker in a different part of the job market from the one they left. A worker who remains employed can receive an outside offer while preserving a current wage and bargaining position. A worker who re-enters after nonemployment may have to evaluate an offer against a different alternative. The transition labels do not by themselves prove how much either mechanism contributes, but they show why continuity belongs in the comparison.
The destination of a direct move also matters. Roughly 25% of women's job-to-job transitions in the study lead to part-time work, compared with about 10% of men's. The headline E-to-E count treats both destinations as job-to-job moves. Yet a part-time destination can change weekly earnings, work hours, and the amount of time available for another job search. It can also represent a valuable improvement in schedule or flexibility. The statistic is therefore not wrong; it is incomplete in a way that matters for both wages and job value.
This is a general rule for interpreting mobility data: a transition rate describes the doorway workers pass through, not the room they enter. To judge whether the move represents progress, we need at least the destination's characteristics and the continuity of the path that produced it.
Starting position changes what growth means
The paper's pay-growth evidence adds a second layer. The researchers compare people who are employed now and were employed one year earlier, then examine the change in their real hourly wages or real weekly earnings. Some of these workers stayed in the same job, some switched employers, and some experienced a period of nonemployment between the two observations. This is not a pure measure of the payoff to a single job switch. It is a measure of how pay changed over a year for a broad group whose starting position can be compared.
At first, the gender difference in growth is close to zero. That result might appear to conflict with the paper's broader conclusion, but it reflects the starting positions of the people being compared. If one group is, on average, lower on a wage ladder, its members may have more room for large gains over the next year. A raw growth comparison can therefore mix two questions: where workers began and how quickly they move from there.
When the researchers control for prior pay, the pattern changes. The female coefficient is −0.110 in the log hourly-wage-growth regression and −0.141 in the log weekly-earnings-growth regression. The paper describes these as roughly 11 and 14.1 log-point gaps in men's favor, conditional on prior pay and the other listed controls. These values are not dollar losses assigned to every woman, nor are they a claim that every switch produces the same difference. They are regression estimates of average conditional growth in the study's sample.
The role of prior pay is easier to see with a comparison between two starting points. A worker earning $20 per hour and a worker earning $40 per hour may each gain $4, but the gains represent different percentage changes and different positions in the distribution. Conversely, two workers can have the same percentage growth while ending at very different wages. Controlling for prior pay does not make the workers identical; it asks whether the gender difference in subsequent growth remains after the measured starting position is taken into account.
This is why “conditional” is not a technical decoration. It changes the question from “who had the larger change?” to “among workers at a comparable measured starting point, who experienced more growth?” The answer in the paper is that women's pay growth is lower. That finding supports the idea of a flatter wage ladder, while still leaving open the mechanisms that generate it.
Prior pay is also an imperfect measure of a worker's starting rung. It captures an important monetary position, but it does not capture schedule control, remote-work access, commuting demands, job security, or bargaining leverage. Two workers with the same hourly wage may therefore begin with different total job values. The conditional estimate sharpens the wage comparison; it does not claim that the regression has made the workers identical in every economically relevant way.
Nor should the estimate be converted into a simple claim that changing employers causes the entire gap. The year-over-year sample includes people who stayed, moved, or passed through nonemployment, and the comparison is observational. The evidence establishes a persistent difference in conditional pay growth in the studied populations. To identify how much comes specifically from switching would require more complete histories of offers, employers, timing, and destinations than this measure supplies.
Hourly wages and weekly earnings also answer different questions. An hourly measure focuses on pay per hour. A weekly measure reflects the earnings associated with a usual work week and can move when hours change as well as when the hourly rate changes. The paper reports both because a labor-market difference can appear in the price of an hour, in the number of hours attached to a job, or in both. The result is not a reason to choose one measure and discard the other. It is a reason to ask which dimension of work each measure captures.
The distinction between individual and occupation-level growth supplies another warning. In the paper's occupation-switch analysis, the conditional gender gap in occupation-level wage growth is about −4.9 log points, while the individual-level wage-growth gap is about −11 log points. The occupation measure is still informative: it shows that women who switch occupations gain less in the average wage attached to the destination occupation, conditional on the earlier occupation wage. But occupation-level wages cannot see every change inside an occupation. An individual can receive a raise, move between firms, change responsibilities, or experience a different wage history while remaining in the same occupation.
An occupation average also compresses workers who are not in identical jobs. The broad BLS categories make this visible at the descriptive level, while the Chicago Fed analysis uses more detailed occupation controls for some regressions. Greater detail reduces some aggregation, but it does not turn an occupation code into a complete record of a worker's employer, tasks, negotiations, or amenities. The difference between the −4.9 and −11 estimates is therefore not a contradiction. It is evidence about what is lost when individual movement is represented by an occupation-level measure.
That loss changes the kind of explanation the statistic can support. An occupation-level result can tell us that the typical wage associated with a destination occupation changed less for women under the comparison used. It cannot tell us whether the missing growth occurred through raises within the same occupation, movement between firms, different job responsibilities, or an unmeasured amenity exchange. Those possibilities are not interchangeable, and the estimate should not be used to choose among them without additional evidence.
A stronger test would follow the same people through more complete employer histories and measure the characteristics of the offers they received and accepted. It would also observe amenities and responsibilities within occupations, not only the average attached to an occupation code. Such evidence could change the interpretation of the gap's mechanism even if it did not change the descriptive finding that conditional individual pay growth differs. Knowing what a measure cannot see is part of knowing what it can establish.
The evidence has a further boundary. The CPS rotation design makes it impossible to observe every intervening job switch with complete accuracy. The study uses a recent-hire definition and robustness checks, but the year-over-year comparison still should not be read as a randomized test in which researchers assigned workers to employers. The SCE contributes detailed information on search and preferences, while the CPS supplies a much larger repeated sample; these sources complement one another without becoming a single dataset with one common denominator.
A careful reading therefore proceeds in layers. First ask who is in the denominator and what transition is being counted. Then ask where the worker went and whether employment continued. Finally compare the outcome with the worker's starting position and choose a measure that matches the question. This sequence turns an apparently puzzling pair of findings—similar job-to-job rates and different pay growth—into a coherent measurement problem.
The conclusion is not that direct job moves do not matter. They remain an important route through which workers can find better matches and gain pay. The conclusion is that the route cannot stand in for the destination or the return. Once those are separated, the remaining question is what makes one destination more valuable than another, even when the cash wage does not rise as much.