What the Research Says

Evidence

A curated collection, grouped by the question each study answers rather than by date, and including the parts of the literature that are not about benefits at all. If you are looking for the short version, the claims and what stands behind each one are on Why Run One.

Two things are worth knowing before the list. First, the political science literature on simulations is small, and a great deal of what is confidently said about them rests on one or two studies or on nothing at all. Second, the large effect sizes people cite come from medical education, where the scenarios are short, the outcomes are technical, and the participants are professional trainees. Neither of those facts is a reason not to run simulations. They are a reason to promise only what the evidence supports.

---

A curated collection

The best in-discipline evidence

Start with the question the field has actually settled. Krain and Lantis (2006), in International Studies Perspectives 7(4), found that a global problems summit produced a different kind of knowledge rather than more of it. Raymond (2010), in the same journal, 11(1), found no significant difference in exam performance and is unusually candid about its own null result. Powner and Allendoerfer (2008), 9(1), reached a similar conclusion with a different design. For twenty years the answer was that a simulation does not teach content better than a lecture does.

Shanks and Zhang (2024), in the Journal of Political Science Education 20(2), is the first serious challenge to that. It is a natural experiment run over three years in an introductory international relations course: some sections were taught with simulations and others with debates and discussions as the comparison, 244 students in all, with mode of delivery controlled. The outcomes are weekly quizzes and a final exam rather than a survey, and the gains held for stronger and weaker students alike. One study at one institution does not overturn three, but it is better powered than any of them and it is the citation to watch.

What simulations do buy is retention. Wunische (2019), JPSE 15(1), is the cleanest statement of the trade: lecture students did better immediately, simulation students retained more over time. Levin-Banchik (2018), JPSE 14(3), is the single most useful citation available for a political science classroom, because the measure is delayed and the setting is a real course. Four topics taught by four methods, then an anonymous quiz three months later: simulation plus an in-class debrief was most effective, simulation alone second, no simulation last. Bernstein and Meizlish (2003), Simulation and Gaming 34(2), found nothing measurable at the end of a congressional simulation course, real differences three years later, and no difference in political participation. Cite that third finding as carefully as the second.

Interest, excitement and self-efficacy

Hendrickson (2021), JPSE 17(2), rebuilt an American Foreign Policy and National Security course around three active components, simulations, a debate and counterfactual analysis, and surveyed students on each one separately rather than on the course as a whole. Excitement, interest and belief in their own ability to succeed all rose. The separate-component design is the useful part, because it shows which activity did the work. Clark and colleagues (2017), JPSE 13(2), found the same direction for interest in European Union politics specifically, and Kensicki and colleagues (2022), JPSE 18(4), found self-efficacy and self-rated communication rising across two weeklong Model G20 summits.

The honest complication is that the largest study in this group did not find it. Clark and Scherpereel (2024), four consecutive years of pre and post surveys across six institutions, found no significant pre/post change in interest in the EU or in politics generally for any group; only discussion of international news moved, and one cohort looks like an outlier dragging the pooled result. So interest is the most commonly reported benefit and it is not the most securely established one.

All of these are self-report, which is not a disqualification but is a limit. Students are reasonably reliable about whether they were interested and unreliable about whether they improved.

Who benefits, and who does not

This is the sharpest material in the literature and the part most often skipped. Engel, Pallas and Lambert (2021), JPSE 17(4), is the one to read if you read only one: in a Model United Nations course, women participated less relative to their numbers, the gap cost them grades, they enjoyed it less, felt less included, and reported less confidence gain. A short intervention designed to close the gap backfired, which the authors read as stereotype threat. Coughlin (2013), JPSE 9(3), found the same pattern at a conference where women were in the majority and men gave most of the speeches and shaped the resolutions. Kaftan and Linantud (2025), JPSE 21(4), found that Statecraft raised everyone’s grades and closed no gap.

It does not all run one way. Davesa and Piros (2019), European Political Science 18(3), surveyed roughly 300 participants across two multi-institution EU simulations held in Brussels and found affective outcomes strongest overall, with women and students not from EU member states among the better performers. The studies are not measuring the same thing: the gap findings come from participation counts and grades inside a single course, and this one from self-reported outcomes after a conference. The useful conclusion is that the distribution of benefit is a design question with a real answer in your own room, not a settled fact in either direction.

The design lever that does seem to matter is power. Baranowski and Weir (2010), JPSE 6(3), found that students holding high-power roles reported more understanding and judged the exercise more worthwhile. The practical consequence is to rotate the powerful roles rather than assigning them once.

How long it has to be

This group barely existed five years ago and it may be the most practically useful one here. Leib and Ruppel (2020), European Political Science 19(3), compared three United Nations simulations of very different scale in the same study: a short in-class Security Council exercise, a regional Model UN, and National Model UN delegations, 159 students in the analysis. Only the long, heavily prepared formats reached all of their intended learning outcomes, and the most demanding goals, leading a debate and solving problems under pressure, were reached by only 10 to 25 percent of students even then. Kensicki and colleagues (2022) point the same way from weeklong summits.

Read that against the format pages on this site and the implication is uncomfortable but useful: what separates a simulation that works from one that does not looks like length and preparation rather than which body you simulate. A one-period exercise is still worth running. It is not a small version of a conference, and it should not be sold as one.

How it gets measured, and why most self-reports mislead

Kartal (2021), JPSE 17(sup1), is the methodological warning to cite whenever somebody reports a survey result about skills: students revise their opinion of their former selves downward, which inflates apparent gains. Oberle, Leunig and Ivens (2020), European Political Science 19(3), put a number on the problem. Across 308 secondary students in fifteen three-hour EU simulations in twelve schools, measured knowledge gain correlated only r = .18 with students’ own sense of having learned, and it was the perceived learning, not the measured kind, that predicted the change in attitudes. The same study carries a clean null: students who already held positive views of the EU did not move.

Clark and Scherpereel (2024), JPSE 20(1), is the largest in-discipline test, four hypotheses across 924 responses from 13 institutions between 2015 and 2020. Knowledge was supported, 6.35 to 7.32 on a ten-item battery, though no better than a lecture course; engagement was mixed; skills were not supported; and empathy was tentatively supported through position shifts in an alter-ego exercise.

What the reviews can and cannot establish

Baranowski and Weir (2015), JPSE 11(4), is the canonical skeptical review, and what it reviews is research design. It is not a pooled meta-analysis and it reports no effect size, so do not cite it as though it does. Duchatelet and colleagues (2019), Educational Research Review 27, is a systematic review of role-play and simulation in higher education.

Chernikova and colleagues (2020), Review of Educational Research 90(4), is the real pooled meta-analysis, and it is the source of the figure everyone quotes: g = 0.85 on complex skills across 145 studies. 126 of those 145 studies are medical education. There is essentially no political science in it, so the number is not evidence about a political science classroom and should never be quoted without that sentence attached.

What other fields can and cannot tell you

Borrowing across disciplines is how the inflated claims got into circulation, so each of these carries the field it came from. Tannenbaum and Cerasoli (2013), Human Factors 55(1), is the best evidence anywhere that the debrief is the active ingredient: 46 samples and 2,136 participants, an overall effect of d = 0.67, and facilitated debriefs at d = 0.75 against d = 0.25 for unfacilitated ones. It is industrial and organizational psychology, 28 of its 46 samples are medical, and “performance” means trained task performance rather than anything a course grades. It converges with Levin-Banchik, which is what makes it worth carrying.

Two negotiation studies cut against the instinct that the performance is the point. Druckman and Ebner (2008), Simulation and Gaming 39(4), ran two randomized experiments and found that students who designed a scenario out-learned students who played roles in one, and retained it a week later. Nadler, Thompson and Van Boven (2003), Management Science 49(4), found that observing a negotiation and comparing two cases beat being told the principle, with observers gaining most and least able to say why. Both are design instructions rather than benefit claims.

And three fields independently reached the measurement conclusion this page keeps repeating. Anderson and Lawton (2009), Simulation and Gaming 40(2), asked why objective assessment of business simulations had barely advanced in twenty-five years while perceptions of learning stayed glowing. Kuehn and Joy (2025), Journal of Legal Education 73, synthesize more than sixty studies of experiential legal education, including moot court and mock trial, and state plainly that most of them lack control groups and pre/post measurement; it is the best answer law has, and it is not a causal one. Schnurr, De Santo and Green (2014), Journal of Geography in Higher Education 38(3), found appreciation of negotiation complexity rising after a role-play while self-rated skill and topic interest fell, which is either a real loss or exactly the recalibration Kartal describes.

One more review is worth knowing about because of what it is not. Vlachopoulos and Makri (2017), International Journal of Educational Technology in Higher Education 14, screened 8,859 studies down to 123 across higher education generally, with business and management the most common field rather than medicine. It reports no pooled effect size, so it supports a direction and not a number.

When it goes wrong

Almost nobody publishes a failure, which is what makes Hendrickson (2023) worth the time. It is an APSA preprint rather than a refereed article and should be cited as one. It starts from an attempt to build a course around board game play that did not work, says why, and then reports what did, with data on the Risk mobile app and on Dice Catan. The argument is that a game whose rules students already know is easier to adopt than one you design, and that what makes or breaks it is whether the game was chosen deliberately for a concept rather than for engagement in general.

The debrief

The field says the learning happens here, and asserts it far more often than it tests it. Wedig (2010), PS: Political Science and Politics 43(3), is the only debriefing structure built for a political science classroom. Crookall (2014), Simulation and Gaming 45(4-5), is the strongest statement of the claim, including that running a simulation without a debrief is unethical, but it is an argument in an editorial and not a finding. Sawyer and colleagues (2016) crosswalk every named debriefing method and conclude that no study has objectively compared them. Schwagele and colleagues (2021) searched Simulation and Gaming and found 631 articles mentioning debriefing and eight that study it methodologically.

Two findings from outside the discipline are worth carrying. Fanning and Gaba (2007) watched thirty-six airline crew debriefs and found that instructors talked more than the crews and rarely allowed silence. Rose and colleagues (2002), a Cochrane review, found that single-session psychological debriefing after real trauma did not prevent PTSD, and that at thirteen months a diagnosis was significantly more likely in the debriefed group. That is a different intervention after a real event and the comparison should not be overdrawn, but it establishes that a structured post-event conversation is not automatically benign. The full treatment is on The Debrief.

What it costs to run one

Almost nothing in the literature addresses the organizational side, which is the part that decides whether a simulation happens twice. Sobisch, Scherpereel, Loedel, Van Dyke and Clark (2018), a chapter in Simulations of Decision-Making as Active Learning Tools (Springer), sets out the costs and benefits of organizing a multi-institutional European Union simulation. Shaw and Switky (2018), JPSE 14(4), is the in-discipline citation for emotional intensity and for helping students step out of role, which is a cost of a different kind and a real one.

What nobody has studied yet

Worth saying out loud, because the gaps are where confident claims go unchecked. There is no verifiable political science study of an AI-assisted or AI-opponent simulation. The role-play literature involving AI sits in language teaching and teacher education, so anything said about it in a political science classroom is at present an opinion. Crisis simulations have design write-ups but almost no measurement. And no study has tested whether the debriefing finding, the most robust result in the whole field, holds for a competitive conference rather than a class session.

---

Handout, free to use

Every citation here was checked against publisher or repository records. One correction worth noting, because it circulates: the 2017 EU simulations article is volume 13, issue 2, and at least one secondary source renders it as 13(1). If you find an error, or a study we have missed, please say so.