What the Evidence Actually Shows
The evidence on JITAIs is promising, but its meaning depends on what was compared, what was measured, and how the intervention was defined. This chapter reads two meta-analytic streams through the fit test developed so far: a mental-health synthesis based mainly on intervention-versus-control comparisons, and an objective-health-behavior synthesis built from pre-post changes. The practical value of the comparison is not to crown one estimate as the “real” JITAI effect. It is to show what each body of evidence can support—and where the numbers stop.
The mental-health estimate is modest, but not empty
The most direct mental-health evidence comes from a systematic review and meta-analysis of JITAIs and ecological momentary interventions, or EMIs, for mental-health and psychological-well-being outcomes. It included 23 studies involving 2,563 adults published between 2018 and May 2025. Twenty-two studies used randomized designs, and one used a quasi-experimental design. The review therefore offers a more informative starting point than a collection of app testimonials, while still representing a mixed set of interventions rather than one standardized JITAI.
Across 22 studies with usable between-group data, the pooled standardized mean difference was Hedges’ g = 0.15, with a 95% confidence interval from 0.05 to 0.26 and p = 0.003. Hedges’ g expresses the difference between intervention and control groups in standard-deviation units after adjustment for small-sample bias. A value of 0.15 is small. It means that the average difference favored the intervention, not that 15% of participants were cured, that symptoms fell by 15%, or that every person experienced a noticeable improvement.
The interval matters as much as the point estimate. The lower bound remains above zero in this analysis, but the plausible range is still consistent with a modest effect. When the researchers removed one outlier, the estimate was g = 0.16, with a 95% confidence interval from 0.06 to 0.26. That sensitivity result makes the direction of the average finding more stable, but it does not turn a small effect into a large one. Nor does it identify which option, tailoring variable, or decision rule produced the result.
The low statistical heterogeneity in the post-intervention between-group analysis— I² = 20%—may look reassuring. Statistical heterogeneity measures variation among observed effects beyond what sampling error would ordinarily explain. But low heterogeneity at this level does not erase conceptual variation. The included interventions differed in their aims, options, decision rules, control conditions, and combinations with other therapeutic elements. Two studies can produce similarly sized estimates while testing very different delivery systems.
The right interpretation is therefore narrow: in this evidence set, current JITAIs and EMIs produced a small average advantage over their controls on the selected mental-health outcomes. That is evidence of potential usefulness, particularly for scalable support in daily life. It is not evidence that all real-time interventions are equally adaptive, that a specific app will work for a particular person, or that a digital option can replace professional care when care is needed.
Persistence is a separate question
The review also asked whether effects remained after the intervention period. Nine studies reported follow-up assessments between one and six months. Their pooled follow-up effect was g = 0.65, with a 95% confidence interval from 0.29 to 1.01 and p < 0.001. On its face, that is a moderate average effect. But the follow-up results were highly heterogeneous: I² = 86%. The studies did not behave as though they were estimating one stable persistence mechanism.
The subgroup estimates illustrate the pattern. Four studies reporting one-month follow-up produced a pooled g = 0.92. Five studies reporting three- to six-month follow-up produced a pooled g = 0.45.