Ask a leadership training provider whether classroom development works, and you will be handed a meta-analysis. Ask a coaching firm the same question, and you will be handed a different meta-analysis. Neither will show you both. The comparison that actually matters to anyone who owns a development budget — how the two modalities perform when measured on the same scale — is almost never made in public, because nobody selling one of them benefits from making it.
So make it internally. Two credible meta-analyses give defensible numbers on each side: Lacerenza and colleagues (2017) on leadership training, and Jones, Woods, and Guillaume (2016) on workplace coaching. Both report delta, a standardized effect size expressed in standard deviation units. Setting them beside each other does not produce a verdict. It produces something more useful: a clear view of what each body of evidence can and cannot carry.
What the training evidence shows
Lacerenza et al. pooled 335 independent samples covering 26,573 participants — one of the largest evidence bases available on any organizational intervention. The overall effect was δ = .76 (95% CI [.64, .89]). Broken out across Kirkpatrick’s four criteria, leadership training produced improvements in reactions (δ = .63), learning (δ = .73), transfer (δ = .82), and results (δ = .72).
The headline number is not the interesting finding. The interesting finding is that those four values were not significantly different from one another. The authors conclude that leadership training tends to be equally effective across criteria — which contradicts the familiar assumption that effects decay as you move from the classroom to the job to the balance sheet. In this dataset, they do not.
Two qualifications belong in any honest reading. First, the reactions estimate rests on only 7 samples and 620 participants, and its confidence interval runs from .12 to 1.15 — wide enough that it should carry no argumentative weight. Second, the authors tested for publication bias directly: a trim-and-fill analysis imputed zero studies, and published and unpublished effects did not differ significantly. That is a stronger clean bill of health than most literatures can produce.
The moderator analysis is where the practical value sits. Effects were stronger when programs were preceded by a needs analysis, incorporated feedback, used multiple delivery methods (particularly practice), spaced sessions over time, were held on-site, and were delivered face-to-face rather than self-administered. Those conditions are controllable. A program that meets none of them is not the intervention that produced δ = .82.
What the coaching evidence shows
Jones, Woods, and Guillaume examined 17 studies (k = 17). Their inclusion criteria matter: they looked only at coaching delivered by internal or external coaches, explicitly excluding manager–subordinate and peer coaching. The finding therefore speaks to formal coaching engagements, not to managers holding better conversations.
Coaching had a positive effect on organizational outcomes overall (δ = 0.36). Across specific outcome criteria, the effects were: skill-based δ = 0.28, affective δ = 0.51, and individual-level results δ = 1.24.
That last figure demands restraint. It is the largest number in either study, and it is the one most likely to be lifted into a proposal deck without its context. It rests on a subset of a 17-study pool — an evidence base roughly one-twentieth the size of the training literature. Treat δ = 1.24 as a promising signal that has not yet been stress-tested, not as an established magnitude.
Jones and colleagues report the overall pool as k = 17 but do not break out, in the published abstract, how many studies sit behind each outcome-specific estimate. So the base for the 1.24 figure cannot be stated precisely from the public record — which is itself a reason to hold it loosely rather than a technicality to wave past.
The moderator findings are more provocative than the effect sizes. Effects were stronger for internal coaches than external ones. The use of multisource feedback was associated with smaller positive effects. And there was no moderation by coaching format — face-to-face, blended, and e-coaching did not differ — or by duration, whether measured as number of sessions or longevity of the engagement. Three of those four findings run against the way premium coaching is typically priced and sold.
Why these numbers are not strictly comparable
Before drawing the comparison, name its limits — otherwise the comparison is just a better-dressed version of the vendor claim it replaces.
- The outcome taxonomies differ. Lacerenza et al. use Kirkpatrick’s four levels, where “results” means subordinate and organizational outcomes. Jones et al. use a different scheme, where “individual-level results” means outcomes for the person being coached. The two share a word, not a construct.
- The evidence bases are asymmetric by more than an order of magnitude: 335 independent samples against 17 studies. Precision differs accordingly, and so should confidence.
- The study pools were assembled under different inclusion rules, from different literatures, with different design mixes. Nothing forces the two samples of research to be exchangeable.
The practical consequence: read these as two independent readings taken on a shared instrument, not as a head-to-head trial. Any executive summary that says “training beats coaching, .76 to 0.36” is performing arithmetic neither study licenses.
Reading the pattern
With that constraint in place, three things are visible.
Coaching’s overall effect is more modest than the market implies. δ = 0.36 is a real effect, but it is a small-to-moderate one, and it sits at roughly half the magnitude of training’s overall figure. The coaching industry’s confidence has run ahead of its evidence base — not because the evidence is negative, but because there is comparatively little of it.
The weakest coaching result is the one most often promised. Skill-based outcomes came in at δ = 0.28, the lowest figure in either study. Coaching proposals routinely lead with skill acquisition. The evidence supports affective outcomes — confidence, attitude, self-efficacy — considerably better, at δ = 0.51.
Where coaching earns its cost is results, provisionally. The individual-level results figure is the genuine case for coaching as an investment rather than a benefit. It is also the figure standing on the thinnest evidence. Both statements are true at once, and a buyer who acts on the first without holding the second is not making an evidence-based decision.
One cross-study contrast deserves attention. Feedback strengthened training effects in Lacerenza et al.; multisource feedback was associated with smaller coaching effects in Jones et al. The same input, pointed in opposite directions by the modality it sits inside. Design inputs do not transfer between delivery models on the strength of intuition.
The comparison, side by side
Where each belongs in a portfolio
Training is the instrument for population-scale capability. When the content is known, the target behaviors are definable, and the population is large, training has both the stronger evidence base and the better cost curve. Its design conditions are documented and controllable, which means a poor result is usually a design failure rather than a modality failure.
Coaching is the instrument for individually variable problems, and its evidence is strongest where the target is affective rather than technical. A leader stepping into a mandate they have not held before, carrying a confidence or judgment problem no curriculum can anticipate, is the case the coaching evidence actually supports.
Combining them is justified when three conditions hold at once:
- The change requires both a shared vocabulary across a group and individually variable application. Absent the first, coach alone. Absent the second, train alone.
- You can meet the design conditions the training evidence identifies. Without needs analysis, spacing, practice, and feedback, the training component contributes cost rather than effect.
- The coached population is small enough that per-participant cost is defensible against a modest overall effect. Coaching does not scale on δ = 0.36 economics.
The combination is not justified when coaching is attached to a training program as an engagement sweetener or a retention gesture. That is a procurement decision wearing a development label, and no part of either literature supports it.
Deciding for your own population
Modality fashion is a poor substitute for a decision rule. Four questions do better.
- What outcome are you buying? Name it in the taxonomy — skill, affect, or results. Effect sizes vary more across outcome types than across modalities, and buyers routinely fund one outcome while measuring another.
- What is your n? Effect sizes are reported per participant; budgets are approved per program. A modest per-person effect across 400 people and a large per-person effect across 6 are different investments with different break-evens.
- Can you meet the design conditions? If the answer is no, discount the published effect size before you build a business case on it. You are not buying the intervention that generated the number.
- What evidence would change your mind? Both meta-analyses average across contexts that are not yours. Decide in advance what you will measure, at what level, and what result would tell you the investment did not work.
The evidence does not name a winner, and this article will not either. It shows that the two literatures are not equally mature, that the outcome definitions are not equivalent, and that the most quoted number in the coaching evidence is also its most fragile. A CHRO who holds all three of those facts at once is in a better position than one holding a single effect size from a single vendor — which, for most of this market, remains the default.
Sources
Lacerenza, C. N., Reyes, D. L., Marlow, S. L., Joseph, D. L., & Salas, E. (2017). Leadership training design, delivery, and implementation: A meta-analysis. Journal of Applied Psychology, 102(12), 1686–1718. https://doi.org/10.1037/apl0000241
Jones, R. J., Woods, S. A., & Guillaume, Y. R. F. (2016). The effectiveness of workplace coaching: A meta-analysis of learning and performance outcomes from coaching. Journal of Occupational and Organizational Psychology, 89(2), 249–277. https://doi.org/10.1111/joop.12119