Many energy-balance related behaviours (EBRBs; e.g., active travel, unhealthy snacking) are performed habitually, with little forethought
. Habits are defined as behavioural patterns, based on learned context-behaviour associations, that are elicited automatically upon encountering associated contexts
. Habits are acquired through context-dependent repetition
, and, once formed, are hypothesised to have two effects on behaviour. First, where associated contexts are consistently encountered and remain stable, habit strength will correlate with behavioural frequency. Second, habit will override motivational determinants of behaviour so that, as habit strengthens, the relationship between deliberative intentions and behaviour will weaken. Subsequently, where habits and intentions conflict, behaviour will tend to proceed in line with habit and not intention
[4, 5]. These hypotheses have been empirically well-supported in EBRB determinant studies
There is growing recognition of the importance of habit in EBRB change. Motivation-based interventions may be insufficient to break established dietary or sedentary habits, because people tend to behave in line with their habits even when motivated not to do so
. Effective behaviour change may depend on disrupting the cue-response relationships that support habitual EBRBs. Conversely, establishing habits for health-promoting EBRBs will facilitate behavioural maintenance, by increasing the likelihood of behaviour persisting even where motivation diminishes
[7, 8]. Recent work has sought to model the habit development process
[3, 9, 10], and EBRB interventions explicitly based on habit formation principles are being trialled
Assessing the extent to which EBRBs are habitual requires a practical, reliable and conceptually robust habit measure. The most popular habit measure in the EBRB domain is the Self-Report Habit Index (SRHI;
[4, 12]). The SRHI comprises twelve items reflecting on three proposed characteristics of habit: automaticity (e.g. [‘Behaviour X is something…’] ‘…I do without thinking’), frequency (‘…I do frequently’), and relevance to self-identity (‘…that’s typically “me”’). The SRHI has been found to detect the habit-behaviour association and moderation of the intention-behaviour relationship in EBRB domains
Recent findings have, however, questioned the parsimony of the SRHI
[4, 13–15]. Various SRHI subscales have been used with no apparent losses in reliability
[16–18], suggesting that some items may be redundant. The SRHI may burden participants unnecessarily, which may be especially problematic in EBRB research, in instances where a multitude of determinants are proposed
, multiple habits are measured (e.g. soft drink consumption and TV viewing
), or, in longitudinal research, habits are assessed at several timepoints
. For example, in one weight-loss intervention trial, participants completed the 12 SRHI items in relation to 14 behaviours over three timepoints (504 items in total
). Unsurprisingly, dissatisfaction was expressed with questionnaire length. Such burden can lead to unreliable or incomplete responses, or withdrawal from the study
. Development of a standardised SRHI subscale may allow more ‘participant-friendly’ assessment of energy-balance related habits.
The conceptual basis of the SRHI has also been questioned. Strong reliability coefficients and a single factor structure have been interpreted as support for a conceptualisation of habit based on three identifiable components: automaticity, behavioural frequency and identity
. However, higher numbers of items biases alpha coefficients towards higher values, and factor analyses may be insensitive to potentially distinct factors on which only one item loads (e.g. identity). A more robust analysis, in which the SRHI was supplemented by additional self-identity items, found that the single SRHI identity item loaded on to a separate factor from all other SRHI items
, suggesting that identity-relevance is not a necessary component of a habit.
Moreover, the incorporation of behaviour frequency indicators in the SRHI is problematic when estimating the relationship between habit and behaviour frequency
[4, 14, 15]. Established habits can only be distinguished from frequent intentional behaviours by automatic activation
[1, 23]. Commentators have thus proposed that the effects of habit on action can be attributed to automaticity
[15, 24], and that it is because habits are automatically elicited that habitual behaviour persists in associated contexts, and deliberative tendencies are overridden
[5, 25]. According to this viewpoint, repeated performance is an antecedent (and consequence) of automaticity
[3, 15, 23], and so the contribution of past behaviour to habit should be reflected by the extent to which behaviours are automatically activated. While behavioural frequency items may be needed to distinguish habit from automatic actions which do not develop through repeated performance, this distinction is rarely of interest in EBRB prediction and habit formation or disruption studies. An automaticity measure may therefore adequately capture habit in these settings. Frequency items are also problematic from a practical perspective, because they can incorporate unidentified stable influences on behaviour
, and so can inflate habit-behaviour associations
. It has been suggested that frequency indicators may not be needed to detect a moderating effect of habit-related automaticity on the intention-behaviour relation
. Gardner et al.
 called for an “SRHI subscale which removes frequency and so may permit a truer estimate of the relationship between cue-response association strength and behavioural performance” (p185).
The present study
This paper describes work to identify and test a SRHI subscale based on behavioural automaticity. Identification of an SRHI shortform would have conceptual and practical benefits for EBRB prediction and habit formation studies. Although automaticity-specific SRHI subscales have been used to study EBRBs
[3, 16–18], no attempt has been made to systematically identify automaticity indicators within the SRHI, and so there has been disagreement about which items best capture automaticity.
We used content validity procedures to identify SRHI items consensually agreed by a panel of researchers to capture automaticity. The convergent validity and predictive utility of the resultant automaticity scale was tested using data from two sources. First, corresponding authors of published SRHI studies, identified via systematic review, were asked to re-analyse their findings using the automaticity subset, and these data were meta-analysed where possible. To maximise data availability, data from all behavioural domains were eligible for analysis. Second, two new primary datasets were collected. Previous SRHI studies have been criticised for neglecting contexts in which habit and intention measures conflict
, and potential contextual cues to habitual action
. To assess the utility of the automaticity subset in these settings, one new dataset measured habits (for unhealthy snacking) alongside counterhabitual intentions (to avoid eating unhealthy snacks), and one dataset used habit items worded to include a potential cue (‘drinking alcohol with the evening meal’). Availability of primary datasets in raw form enabled comparisons between the automaticity index and a composite of SRHI items removed from the automaticity subset, which we did not deem feasible to request from authors of published studies. A further two datasets, which formed the basis of previously published work
, were also available to us in raw form and permitted comparisons with an additional habit measure (the transport-specific ‘Response-Frequency Habit Measure’
). Together, the four primary datasets covered both sides of the energy balance equation: energy expenditure (inactive and active commuting) and intake (snacking, alcohol consumption
In all analyses, the automaticity index (which we term the ‘Self-Report Behavioural Automaticity Index’; SRBAI) was assessed against the criteria by which the SRHI has been tested and become established: reliability, convergent validity, and predictive validity. We hypothesised that:
Hypothesis 1. (a) The SRHI and SRBAI (and RFM) will be intercorrelated, and (b) will each correlate with behaviour. However, (because of the removal of items which assess frequency and identity) (c) the automaticity-specific SRBAI will be less strongly correlated with behaviour than will the SRHI, or (d) a scale comprised of SRHI items removed from the SRBAI.
Hypothesis 2. (a) The SRBAI and SRHI will each moderate the relationship between intentions and behaviour, such that where habit is strong, intentions will have a weaker effect on behaviour, and vice versa. Assuming that the moderating effect of habit is attributable to automaticity, then, due to the removal of strong automaticity indicators, (b) a composite of SRHI items omitted from the automaticity subset will fail to detect moderation of the intention-behaviour relationship.
Support for Hypotheses 1a, 1b and 2a would show that the SRBAI can capture habit-behaviour effects to the same extent as the SRHI. Support for Hypotheses 1c, 1d and 2b would suggest that the SRBAI excludes items that may exaggerate true habit-behaviour associations or obscure the expected habit x intention interaction.
Hypotheses 1a, 1b, 1c and 2a were assessed using secondary data and the four primary datasets. Hypotheses 1d and 2b were assessed using the four primary datasets only.