Habits and behaviour change: the research

How habits form: Lally and colleagues (2010)

The most cited study of time to automaticity was published in 2010 by Lally, van Jaarsveld, Potts and Wardle. Ninety-six volunteers each chose an eating, drinking or activity behaviour to perform daily in the same context, for example after breakfast, for 12 weeks. They completed a self-report habit index every day and recorded whether they had done the behaviour. Eighty-two gave enough data for analysis, and the researchers fitted a curve to each person's automaticity scores across the 84 days.^[1]

The model fitted for 62 individuals, and only 39 of those showed a good fit. The authors report that performing the behaviour more consistently was associated with a better fit. These numbers show how modest the dataset is: the headline result rests on a minority of an already small sample, and every measure of automaticity was self-reported. The British Psychological Society's Research Digest describes reliance on self-report as a shortcoming that the researchers themselves acknowledged.^[1][2]

What the 66 days figure does and does not mean

The Research Digest summary of the study states that the average time to reach maximum automaticity was 66 days, varying between participants from 18 days to a predicted 254 days. The upper figure is an extrapolation, because the study ended at 84 days while some participants' automaticity was still rising. The paper's own abstract reports the range as the time taken to reach 95 per cent of each person's asymptote of automaticity, which ranged from 18 to 254 days, and it stresses that this indicates considerable variation between people.^[1][2]

The study also found that missing one opportunity to perform the behaviour did not materially affect the habit formation process. The Digest adds that too many missed repetitions lowered the level of automaticity eventually reached, and that 66 days is much longer than most previous estimates. The figure applies to daily eating, drinking or activity behaviours chosen by volunteers, so it should not be read as a promise for a complex goal. No independent replication was located in the sources reviewed for this page, so the result is best treated as indicative.^[1][2]

Implementation intentions (Gollwitzer)

Implementation intentions are specific if-then plans that link a situation to an action, a concept associated with the psychologist Peter Gollwitzer. Gollwitzer and Sheeran published a meta-analysis in 2006. A 2021 meta-analysis by Wang and colleagues reports that it found a positive effect on goal attainment across achievement, relationship and health goals, a medium to large effect (Cohen's d = 0.65, 95 per cent confidence interval 0.6 to 0.7). This page relies on that summary, because the 2006 chapter itself could not be opened.^[3]

Narrower meta-analyses report smaller figures. According to the same 2021 paper, Presseau and colleagues (2017) found small to medium effects on objectively assessed health behaviour (d from 0.14 to 0.37) and medium effects on self-reported behaviour (d from 0.19 to 0.77). Adriaanse and colleagues (2011) found a smaller effect for reducing unhealthy eating (d = 0.29) than for adding healthy foods (d = 0.51). Wang and colleagues note that the larger 2006 figure came from analyses that included laboratory, related and intervention studies.^[3]

The same paper records qualifications. It cites work finding no promoting effect for highly impulsive individuals (Churchill and Jessop, 2010) and a 2018 meta-analysis by Da Silva and colleagues finding that implementation intentions without reinforcement had no significant effect on adult physical activity. A Cohen's d is a standardised difference between groups and not a percentage gain, so d = 0.65 does not mean that people became 65 per cent more likely to succeed.^[3]

Goal setting and progress monitoring

A 2016 meta-analysis of multi-component goal setting interventions for physical activity pooled 45 articles (n = 5,912) and reported a medium positive effect (Cohen's d = 0.552, 95 per cent confidence interval 0.43 to 0.67). Harkin and colleagues' 2016 meta-analysis of 138 randomised studies (N = 19,951) found that interventions to promote monitoring of goal progress increased how often people monitored (d+ = 1.98) and promoted goal attainment (d+ = 0.40, 95 per cent confidence interval 0.32 to 0.48). Effects on attainment were larger when outcomes were reported or made public and when progress was physically recorded.^[4][5]

Results are not uniform. A 2019 systematic review of goal setting in geriatric rehabilitation found no statistically significant difference from usual care for physical functioning (standardised mean difference -0.11, 95 per cent confidence interval -0.32 to 0.10), and the authors rated the evidence low quality with a high risk of bias. Psychology also has a wider replication problem: the Open Science Collaboration's 2015 attempt to repeat 100 studies found replication effects half the size of the originals, which is a reason to treat early single estimates with caution.^[6][7]