Evidence
Tiny Shifts is built on a narrow, real core of evidence about what happens between sessions. This page shows the findings we lean on, the limits we hold ourselves to, and every study behind them.
Six findings that shape the product. Each names its study; the numbers are the authors' own.
The one causal finding
Across 46 studies, adding between-session tasks to the same therapy improved outcomes, with a controlled effect size of d = 0.48. It is the only finding here that rests on a causal design, and it is why the loop is assign, carry, and review, not send and forget.
Kazantzis, Whittington and Dattilio, 2010. Clinical Psychology: Science and Practice. 46 studies, 1,072 participants.
The working relationship
In coaching, the working alliance correlates with outcome at r = .41 across 27 studies, and holds for every desirable outcome category. In psychotherapy the figure is r = .28 across 295 studies. This is why nothing reaches your client without you, and why every message carries your name.
Graßmann, Schölmerich and Schermuly, 2020, Human Relations, 3,563 coaching processes. Flückiger, Del Re, Wampold and Horvath, 2018, Psychotherapy, more than 30,000 patients.
If-then plans
Plans of the form "when X, I will Y" help people act on intentions. Across 94 studies the headline effect is d = 0.65, but field studies with self-administered plans run closer to g = 0.24 to 0.33. So the messages name a moment and a cue in the client's own words, and we do not promise more than the field numbers.
Gollwitzer and Sheeran, 2006, Advances in Experimental Social Psychology. Wang, Wang and Gai, 2021, Frontiers in Psychology.
Intentions alone
People act on their intentions roughly 53% of the time, and experimentally changing intentions explains only about 3% of the variance in what people then do. A product that stops at goal-setting and reminding should be expected to under-deliver, which is why the client's replies and the next session are part of the loop.
Webb and Sheeran, 2006, Psychological Bulletin, 47 experimental tests. Sheeran, 2002.
Feedback to the practitioner
Feedback systems improve outcomes by alerting the practitioner, not by giving the client a dashboard. The large 2018 IAPT trial found no overall effect, and a real one for clients who were not on track. That is why replies come back to you before the next session, and why the client never sees a score.
de Jong and colleagues, 2024, 25 years of progress feedback research. IAPT outcome feedback trial, 2018.
Completion
Completion of between-session work correlates with outcome at r = .26 across 23 studies. When completion is logged objectively, as software does, the figure is r = .16 to .22. We quote the lower number and we do not claim that more messages mean more change.
Mausbach and colleagues, 2010, Cognitive Therapy and Research, 2,183 participants. Kazantzis and colleagues, 2016, Behavior Therapy.
The limits we hold ourselves to. Every one of these is written into how we talk about the product.
Only the between-session work contrast comes from a controlled design, and even that confounds the task with the practitioner's attention. Nothing here licenses "if we increase X, outcomes improve by Y".
The strong evidence comes from CBT with depression and anxiety samples. Coaching is exactly where the evidence is weakest, so importing those magnitudes to a coaching practice is an extrapolation, and we say so.
Uncontrolled before-and-after coaching studies show roughly three times the effect of controlled ones, g = 1.15 against g = 0.39, and most rely on self-report. We quote the controlled figures.
Objective logging, field settings, self-administered delivery and controlled designs all shrink the headline numbers. A text-delivered, objectively logged product sits on the low side of each one.
How often to message, how fast people habituate, and whether scaffolded follow-through creates dependence once prompting stops are open questions. That is what the three-month design partnership is for.
The alliance holds up in guided online treatment, r = .28 across 23 samples. That evidence covers practitioner-delivered online care, not asynchronous prompts, so we treat alliance protection as the design constraint, not a solved problem.
Every source behind the findings above, with a link to the paper.
Clinical Psychology: Science and Practice
Kazantzis, Whittington and Dattilio, 2010
46 studies. Controlled same-therapy contrast, with vs without homework: d = 0.48.
Read the studyCognitive Therapy and Research
Mausbach, Moore, Roesch, Cardenas and Patterson, 2010
23 studies, 2,183 participants. Pooled r = .26; objectively measured completion r = .16 to .22.
Read the studyBehavior Therapy
Kazantzis and colleagues, 2016
17 studies, 2,312 participants. Quantity of homework completed, g = 0.79 at post-treatment.
Read the studyAdvances in Experimental Social Psychology
Gollwitzer and Sheeran, 2006
94 studies. d = 0.65 overall; the field estimate for self-administered plans is about a third to a half of that.
Read the studyFrontiers in Psychology
Wang, Wang and Gai, 2021
Realistic magnitude for self-administered text or email delivery: g = 0.24 to 0.33.
Read the studyHuman Relations
Graßmann, Schölmerich and Schermuly, 2020
27 studies, 3,563 coaching processes. r = .41, positive across every desirable outcome.
Read the studyPsychotherapy
Flückiger, Del Re, Wampold and Horvath, 2018
295 studies, more than 30,000 patients. r = .278; r = .275 in internet-delivered treatment.
Read the studyThe Journal of Positive Psychology
Theeboom, Beersma and van Vianen, 2014
Controlled designs g = 0.39 against uncontrolled g = 1.15. The source of our controlled-only rule.
Read the studyPMC, 2024
de Jong and colleagues, 2024
Feedback improves outcomes by alerting the practitioner; the effect concentrates in not-on-track clients.
Read the studyIAPT outcome feedback trial
Pre-print, 2018
No overall effect of feedback; a real effect, d ≈ 0.2, for clients who were not on track.
Read the studyAdministration and Policy in Mental Health
Systematic review, 2015
The practitioner loop carries the effect; client-facing surfaces earn their keep on engagement.
Read the studyU.S. Bureau of Labor Statistics
Full-time employed adults
The week behind the 168 numeral on the home page: one cell per hour, one hour the session, 167 without you.
Read the studyTwo or three design partners. Three months with real clients. What we learn about frequency, habituation and follow-through comes from you.