Do Mentorship Programs Work? What the Evidence Says

Evidence grade: Mixed. Credible causal studies exist and disagree, or the effect holds only under conditions the studies can name. This grade describes the published evidence. It is not a prediction about your workforce and not a result RetainScore has produced.

Two studies that belong in every mentoring deck, and rarely appear

Mentoring is one of the few retention interventions with a large randomized trial behind it. The trial is the reason this grade is not insufficient, and its result is the reason it is not strong.

The randomized trial found nothing on retention

In 2005 the U.S. Department of Education's research arm randomized 418 elementary schools across 17 urban districts, covering 1,009 beginning teachers, into a comprehensive induction program or business as usual. The program was not a light touch: "Beginning teachers in treatment schools were then assigned to a full-time mentor with a 12 to 1 ratio", with trained mentors, monthly professional development and time observing veteran teachers. Some districts ran it for one year, some for two. The teachers were followed for four years.

The finding on the outcome mentoring programs are bought for: "There was no impact on teacher retention over the first four years of the teachers' careers". The report is specific about how thoroughly that null held: "This was true of retention in the original school, the original school district, and the teaching profession". Sensitivity analyses for survey non-response did not change it.

The program was not inert. In the districts that ran it for two years, students of treated teachers scored higher in the third year, enough to "move the average student from the 50th percentile up 4 percentile points in reading and 8 percentile points in math". A well-resourced mentoring program can improve how people do the job. That is a different claim from keeping them in it.

The long-run study found mentored officers stayed

The strongest result on the other side comes from the Army. Payne and Huffman surveyed "Over 1,000 U.S. Army officers responded to two surveys over a two-year period" and then looked at who had actually left a decade on. Officers who had a mentor were less likely to have gone, and part of that ran through how attached they felt to the organization: "Affective commitment partially mediated the negative relationship between mentoring and actual turnover behavior ten years later".

Two things about that study limit what it can carry. It is observational: officers who had mentors are compared with officers who did not, and the design cannot separate being mentored from being the kind of officer who attracts a mentor. And it measures having a mentor, not enrolling in a program. Those are not the same thing, and the gap between them is where most of the practical advice below comes from.

What the meta-analysis says

Eby and colleagues pooled 112 studies and 116 independent samples across youth, academic and workplace mentoring. Their summary sentence is short: "All of the effect sizes were small in magnitude". Intentions to leave were modestly lower among mentored people across all settings. Actual leaving is the outcome that matters here, and for workplace mentoring it did not move: "Both youth and academic mentoring were significantly associated with withdrawal behavior, whereas workplace mentoring was not".

The authors are careful about what the correlations mean. Because most of the underlying studies are cross-sectional, "it is unknown whether significant correlations between mentoring and outcomes reflect a causal effect of mentoring". Their own recommendation is to "temper what are sometimes seemingly unrealistic expectations about what mentoring can offer".

When it does seem to move retention

Rockoff's study of New York City's 2004 mentoring program is the one that says something about design rather than just direction. Using instrumental variables and school fixed effects to get around the fact that mentors were not randomly assigned, he found that measures of mentoring quality predicted what teachers said about their mentors far better than what they did next: "weaker evidence of effects on teacher absences, retention, and student achievement".

One retention result did survive: "retention within a particular school is higher when a mentor has previous experience working in that school". More hours of mentoring also mattered, at least for students: "student achievement in both reading and math were higher among teachers that received more hours of mentoring". Matching mentors to teachers by subject, the thing program rules usually insist on, showed little effect.

Put the two teacher studies together and a pattern appears. The nationally designed program with external mentors did nothing for retention. The local program's one retention effect ran through mentors who knew the specific workplace. What seems to keep a new hire is someone who can explain how this place actually works, not the existence of a mentoring scheme.

Why the grade is mixed

Mixed is the grade for credible studies that disagree, or for an effect that holds only under conditions the studies can name. Both apply. The randomized trial and the long-run observational study point in different directions, and the one design finding that survives, local knowledge, is a condition rather than a general effect.

The honest boundaries

• Teachers and army officers. The causal and long-run evidence comes from two occupations with unusual career structures. A software team or a care home may differ, and no study of comparable quality exists for them.
• Programs are not mentors. The study with the positive result measured whether someone had a mentor. The randomized trial tested a program that assigned one. Assigning a mentor does not reliably create the relationship the positive studies describe.
• Small effects, if any. Even where associations are significant, the meta-analysis finds them small. A mentoring program is not a large lever on quits.
• Be skeptical of retention percentages attached to mentoring. The most-quoted figures in vendor decks come from company case studies with no published method. A before-and-after inside one firm cannot separate the program from the hiring market, the manager, or the fact that people who opt into mentoring were already more likely to stay.

If you run one anyway

• Pair people with someone who knows this unit. That is the only design feature with a retention result behind it.
• Protect the hours. Time with the mentor was the input that moved outcomes in the New York data. A named mentor who is never available is a line on an org chart.
• Measure against a comparison group. Mentored employees stayed longer is what selection looks like. Compare against similar hires who were not offered the program, or against the same roles before it existed.
• Do not stop looking for the cause. If new hires are leaving, the mentor is downstream of whatever is pushing them out. The retention diagnostic is built to find that first.

Read the companion grade on whether structured onboarding works, which covers the same first months from the other side and has the best evidence on the site.