Why do new members stop attending health and itness venues? The importance of developing frequent and stable attendance behaviour of

S Objectives: Attendance at health and itness venues often declines dramatically during the irst year of mem- bership. The present study sought to identify factors associated with continued attendance in new members. Methods: A secondary data analysis was conducted on the attendance data of 1726 new members of a health and itness organisation over the irst 12 months of their membership. Associations were assessed between members ’ frequency and pattern of attendance during the irst quarter of their membership, age, gender, and home location and their attendance levels in the fourth quarter after joining. Results: There was a signiicant reduction in the frequency of attendance over time from a mean of 7.48 times in the irst month to a mean of 0.92 in the 12th month after joining. Older age, starting membership in the autumn or spring, frequency of attendance in the irst quarter and stability of the context of attendance (i.e., same time and location) in month three were all signiicantly associated with increased attendance levels in the fourth quarter. Conclusions: The indings indicate that new members ’ early attendance behaviour, in terms of the frequency and the stability of attendance, may be important for supporting continued attendance at health and itness venues. Interventions to increase longer-term attendance and, in turn, physical activity, should focus on promoting regular and consistent attendance, especially in terms of day, time and location, early on in individuals ’ mem- bership of health and itness venues.


Introduction
Physical activity (PA) is associated with many health beneits including the prevention of chronic conditions such as cardiovascular disease, cancer and type 2 diabetes (Albright et al., 2000;Clague & Bernstein, 2012;Whelton, Chin, & Xin, 2002). Public Health England (PHE) recommends that adults undertake at least 150 min of moderate intensity activity each week in bouts of 10 min or more or 75 min of vigorous intensity activity spread across a week (PHE, 2016). However, in the UK it is estimated that 26% of all adults do not meet the recommended guidelines for PA in a week (Sport England, 2016). Increasing the levels of PA in the adult population remains a public health priority.
According to PHE, the sport and leisure sector is one of four broad areas, along with local authorities, National Health Service commissioners and healthcare professionals, where local and national bodies can take action to ensure people are active (PHE, 2016). However, the sport and leisure sector is the only one that provides PA as a primary service. Health and itness is one of the largest subsectors of the sport and leisure industry, with approximately 15% of the population estimated to be a member of a health and itness venue in the UK (The European Commission, 2014). Health and itness venues typically provide PA equipment within gyms, and many offer swimming pool facilities and exercise class activities. These locations are also where many PA professionals (e.g., personal trainers in gyms, swimming teachers and PA class instructors) are employed (REP, 2019). Health and itness venues are therefore ideally placed to help to increase population PA levels. However, little is known about the extent to which people actually use these venues.
A study of 259,355 ex-members at 267 Dutch health and itness venues concluded that only 10% of members attended regularly (deined Abbreviations: PA, physical activity; PHE, Public Health England. as at least four times a month) for the irst six months of their membership (Middelkamp, Van Rooijen, & Steenbergen, 2016). Similarly, a study in the United States of health and itness members' attendance concluded that many members were 'paying not to go to the gym' (Dellavigna & Malmendier, 2006). In particular, the mean attendance of 145 members on an annual contract was only 4.36 times a year. Given that many members join a health and itness venue to improve their health or physical itness (Crossley, 2006), these indings are concerning from a public health perspective. Moreover, little is known about the factors that are associated with continued (or discontinued) attendance at health and itness venues.
Several factors have been associated with maintenance of PA participation, more generally, that might also be expected to be associated with continued attendance at health and itness venues, including younger age, male gender, seasonality, and proximity to a sports facility (Cepeda et al., 2018;Marques, Martins, Peralta, Catunda, & Nunes, 2016;Rich, Grifiths, Dezateux, 2012). In addition, past behaviour is one of the strongest correlates of PA, with an average correlation of r = 0.54 across a meta-analysis of 44 studies (McEachan, Conner, Taylor, & Lawton, 2011). A key way in which past behaviour may impact on continued PA is through the formation of a habitual response by repeatedly performing the same behaviour in the same context. Habit is a process formed through repetition of behaviour in a speciic context (Lally, van Jaarsveld, Potts, & Wardle, 2010) in which a stimulus generates an impulse to act, resulting in a learned stimulus-response association (Gardner, 2014). Habitual behaviour is therefore proposed to develop through repeated execution of behaviours in the consistent presence of salient cues or contextual features (Gardner & Lally, 2018;Wood, 2017). Thus, both the frequency with which, and the stability of the context in which, the behaviour is performed are key antecedents for forming a strong habit. Moreover, Ouellette and Wood (1998) have argued that as behaviours are repeated more often in stable contexts, their performance switches from being under the control of intentional (i.e., relective) to habitual (i.e., automatic) processes. In line with this argument, Norman and Cooper (2011) reported that a measure of habit strength was only associated with subsequent behaviour (breast self-examination) when it had been performed frequently in a stable context (i.e., same time and place) in the past. Similarly, Kaushal and Rhodes (2015) reported that engaging in regular exercise for at least six weeks in a consistent context (i.e., place and time) was predictive of habit formation. Measures of habit strength have, in turn, been found to have a medium-sized average correlation with PA (r = 0.43) (Gardner, De Bruijn, & Lally, 2011). In addition, a systematic review found that 26 of 37 studies included in the review reported a signiicant positive association between measures of habit and PA (Rebar et al., 2016). The development of habitual behaviours has also been proposed to have an impact on increased attendances in health and itness venues (Calzolari & Nardotto, 2017;Muller & Habla, 2018).
The aim of the present study was to analyse the pattern of new members' attendance behaviour at a health and itness organisation's venues and to assess potential correlates of continued attendance, including age, gender, proximity to a venue, seasonality, frequency of early attendance and context stability (i.e., time and location of attendance). In particular, the following research questions were addressed: (1) What are the attendance levels of new members of a UK health and itness organisation over the irst 12 months of their membership? and (2) What are the correlates of continued attendance?

Methods
This study reports a secondary data analysis of attendance records of new members at a health and itness organisation over a 12-month period.

Design
Attendance data of individuals who joined a health and itness organisation as members between October 2015 and July 2016 were included for analysis. Anonymised data were extracted from the health and itness organisation's database.

Participants and study setting
All included members had purchased a 12-month contract with the health and itness organisation. Such contracts cannot be cancelled during their duration. The organisation has six venues in a UK city, each offering gym equipment and group exercise classes, with ive of the venues providing swimming pool facilities. Individuals who utilised the health and itness facilities on a 'pay as you go' basis were excluded as were individuals who were part of exercise referral schemes. In total, 1726 new members were included in the sample for data analysis.

Procedure
Ethical approval for the study was obtained from the School of Health and Related Research, University of Shefield Research Ethics Committee. Following ethical approval, anonymised attendance data of the health and itness members were downloaded from the organisations' reporting system. The data were then cleaned to remove members who did not meet the inclusion criteria. Attendance records for each member were then automatically counted and categorised into four week periods to produce 12 months of attendance data (with each month consisting of four weeks). All statistical analyses were conducted in SPSS (version 25).

Measures
Each time a member uses one of the organisation's health and itness venues, the location, date and time of the attendance is electronically recorded through their member card or wristband at the venue turnstile. This automatic recording of attendance is saved on a database. This electronic record of attendance was used for the analysis in the current study. Using these data, the frequency of attendance in each month and quarter was analysed. In addition, a measure of context stability was calculated from the number of times a member attended the same venue on the same day of the week and within a 3-h time window more than once during a month. Thus, if members attended the same venue on the same day of the week and within a 3-h time window twice during a month they had a context stability score of 1, three times they had a score of 2 and four times they had a score of 3. If members did not attend the same venue on the same day/time more than once in a month, they received a score of 0. The total of these scores in month 3 was taken as the measure of context stability.
Age, gender and member start date were also recorded in the database and included for analysis. The distance in miles members lived from a venue was calculated from the irst part of their postcode and used as a measure of location in the analyses.

Statistical analysis 2.5.1. Frequency of attendance
Two repeated measures ANOVAs were conducted to assess changes in the mean frequency of monthly and quarterly attendance over the 12month period. The percentage of members attending at least once a month over the 12-month period was also calculated.

Correlates of attendance
First, associations were examined between each of the independent variables (i.e., age, gender, location, frequency of attendance in the irst quarter, context stability, and season) and frequency of attendance in the fourth quarter. Pearson's correlations were conducted to assess how strongly age, location, context stability in months one to three and frequency of attendance in the irst quarter were associated with frequency of attendance in the fourth quarter. Independent t-tests were conducted to assess associations between gender and frequency of attendance in the fourth quarter. An analysis of variance was conducted to assess the effect of the season of members' start date on attendance in the fourth quarter. Second, a multiple regression analysis was conducted to examine how much variance these variables could explain the frequency of attendance in the fourth quarter as well as the unique contribution of each independent variable.

Power analysis
A power analysis indicated that with a sample of 1,726, it would be possible to detect small-sized correlations (r = 0.07), according to Cohen's (1992) effect size indexes, between the potential correlates of continued attendance and frequency of attendance in the fourth quarter, at 80% power and alpha set at 0.05.

Participant characteristics
Of the 1726 members, 845 (49%) were female and 881 (51%) were male. Members ranged in age from 19 to 70, with a mean age of 35.30 (SD = 10.21). There was missing data on age for 20 members. Members resided in 52 different postcode areas, with members living on average 3.4 miles from the nearest health and itness venue. There was missing data on location for 15 members.

Frequency of attendance
The mean attendance for each month is reported in Table 1. In total across the 12 months, members attended the venues 59,762 times. Over the 12 months there was a signiicant effect of time on attendance, F(11, 1715) = 260.61, p < .001. Post-hoc comparisons using Bonferroni posthoc tests indicated there were signiicant differences (p < .05) between each of the months apart from between months eight and nine (p = .418). As shown in Figure 1, each month had a drop in attendance from the previous month. The largest drop in mean attendance was between months one and two, t(1725) = 63.68, p < .001. There was also a signiicant effect of time on attendance over the four quarters, F(3, 1723) = 688.66, p < .001 (see Table 2). Post-hoc comparisons using Bonferroni post-hoc tests indicated that each quarter was signiicantly different from each other (p < .01). The drop in attendance can also be seen in the percentage of members who attended at least once each month over the irst 12 months of their membership. As shown in Table 1, 100% of members attended at least once in the irst month of their membership; this igure fell to 50% and 22% at six and 12 months.

Correlates of attendance in quarter four
Considering the correlates of attendance frequency in quarter four, there was a small but signiicant correlation between age and attendance in quarter four, r(1724) = 0.11, p < .001, such that older members attended more frequently. Neither gender, t(1724) = 1.52, p = .13 d = 0.07, or distance to the nearest venue (i.e., location), r(1724) = 0.009, p = .71, were signiicantly associated with attendance in quarter four. Season had a signiicant effect on attendance frequency in quarter four, F (2, 1722) = 508.43, p < .001, η p 2 = 0.016. Post hoc analyses indicated that members who started their membership in autumn (M = 4.52, SD = 8.54) or spring (M = 4.55, SD = 8.84) attended more frequently in quarter for than those who started in summer (M = 2.35, SD = 6.13) or winter (M = 2.88, SD = 6.26). There was a signiicant medium-sized correlation between frequency of attendance in quarter one and attendance in quarter four, r(1724) = 0.36, p < .001, indicating that increased attendance in quarter one was associated with increased attendance in quarter four. Context stability in month one, r(1724) = 0.19, p < .001, month two, r(1724) = 0.28, p < .001, and month three, r (1724) = 0.36, p < .001, were signiicantly correlated with attendance in quarter four, such that increasing levels of context stability in each of the irst three months were associated with increased attendance in quarter four. The strongest (medium-sized) correlation was between context stability in month three and attendance in quarter four.

Regression analysis to predict attendance in quarter four
A regression analysis was conducted with age, gender, location, season (autumn/spring versus summer/winter), frequency of attendance in quarter one and context stability in month three as the independent variables and frequency of attendance in quarter four as the dependent variable. The regression model was signiicant, F(6, 1684) = 59.20, p < .001, f 2 = 20, and explained 17% of the variance of attendance behaviour in quarter four; a medium-sized effect according to Cohen's (1992) indexes. Table 3 displays the results of the regression analysis. Inspection of the beta weights revealed that age, season, attendance in quarter one, and context stability in month three made signiicant unique contributions to the prediction of attendance in quarter four. Thus, increased frequency of attendance in quarter four was associated with older age, starting membership in autumn or spring, increased quarter one attendance, and increased context stability (i.e., attending at the same place and time) in month three.

Discussion
This study analysed the attendance data of new members of a health and itness organisation in the UK over the irst 12 months of their membership. Substantial and signiicant reductions were found in attendance over time, with the mean frequency of attendance falling from 7.48 in month one, to 2.44 in month six and 0.92 in month 12. Similarly, the percentage of new members who attended at least once fell from 100% in month one, to 50% in month six and 22% in month 12. The levels of attendance in the current study suggest that by the end of the irst year of membership, many members do not attend health and itness venues often. This inding is in line with previous studies in the Netherlands and USA (Dellavigna & Malmendier, 2006;Middelkamp et al., 2016). Given that many members decide to join a health and itness venue in order to increase their levels of physical activity to address concerns about physical health or itness (Crossley, 2006), it is unlikely that the low level of attendance is due to members engaging in PA in locations outside of the health and itness venue. As a result, it is likely that most of the members who stop attending remain physically inactive.
Considering the correlates of attendance, older members were found to attend more often in quarter four than younger members, although the size of the effect of age on attendance was small according to Cohen's (1992) criteria. The signiicant positive correlation is in contrast to the majority of previous research on PA which has found that activity decreases with age (Rhodes et al., 1999). However, the mean age of the members in the current study was only 35.30 and only 1.3% of the sample was aged over 60. It is possible that younger members in the present sample may have faced increased barriers in terms of other commitments or competing priorities (Nikolaou, Hankey, & Lean, 2015;Strong, Parks, Anderson, Winett, & Davy, 2008) which may have reduced attendance. In addition, the signiicant correlation between attendance and age might also be explained by the important social factors for engaging in PA in older people (Franco et al., 2015); these factors are likely to be present in health and itness venues (e.g., interaction with peers or dependence on professional instruction). The inding that males and females had similar levels of attendance is in contrast to previous research on PA which has identiied males to have higher PA rates than females (Althoff et al., 2017). The current inding is somewhat surprising given that various factors, such as self-eficacy, social support, and motivation have been previously identiied to impact on differences in PA participation between males and females (Edwards & Sackett, 2016). In addition, barriers such as time expenditure or childcare responsibilities have also been identiied to be factors in non-exercising adult females (El Ansari & Lovell, 2009). It is possible that membership of a health and itness venue may help to overcome some of these barriers. For example, venues may provide personal support and encouragement to exercise and members may be able to it attendance in and around other commitments (e.g., before/after work or during a lunch break).
The inding that those living closer to a health and itness venue were no more or less likely to attend a health and itness venue in quarter four also contrasts with previous research outlining the importance of accessibility and proximity as a correlate of PA (Sallis et al., 2016). One possible reason for the current inding may be the small number of members (3.4%) living in the same postcode location as the most frequently attended venue which is located in the city centre. It is possible that many members attend this venue when they are already near the venue location (e.g., for work). Unfortunately in the current dataset there was no record of the location where members worked; future research could seek to assess accessibility and proximity in relation to both members' home and workplace.
The inding that members who joined the health and itness organisation in autumn or spring attended more often in quarter four than those who joined in winter or summer is consistent with previous research that has reported seasonal variations in PA (e.g., Cepeda et al., 2018;Rich, Grifiths, & Dezateux, 2012). There are two possible explanations for why members who joined in the summer or winter were less likely to maintain their attendance. First, the reasons associated with joining in winter (e.g., New Year's resolutions) and summer (e.g., wanting a 'summer body') may not be strong enough to maintain continued attendance, especially if the expected outcomes are not obtained (Rothman, 2000). Second, there may be more disruptions to people's normal routines in winter (e.g., bad weather) and summer (e.g., extended holidays) that prevent the formation of strong habits that help to maintain new behaviours (Kwasnicka, Dombrowski, White, & Sniehotta, 2016).
Frequency of attendance in the irst quarter was associated with frequency of attendance in the fourth quarter. This inding is in line with previous research that has shown that past behaviour is a strong correlate of future PA behaviour (McEachan et al., 2011). There are two main ways in which past behaviour is hypothesised to inluence future behaviour (Ajzen, 2011). First, past behaviour may inluence people's beliefs about the behaviour which, in turn, inluences their decisions. For example, members may ind attending rewarding or enjoyable and make a conscious choice to keep attending as a result. Accordingly, in a   Note. **p < .01, ***p < .001. study of 94 health and itness members, cognitions from the theory of planned behaviour predicted maintenance of PA behaviour over a 12-week period (Armitage, 2005). Second, when a behaviour is performed frequently in a consistent context it can then be performed relatively automatically with little conscious deliberation (Ouellette & Wood, 1998). For example, members may attend venues to do a particular class at the same time each week. In the current study, measures of both the frequency of attendance and the extent to which to which members attended at the same venue/ time (i.e., context stability) were found to have signiicant, and mediumsized, effects on attendance in the fourth quarter. These indings are consistent with the idea that frequent and stable attendance behaviour in the irst few months of membership may have led to the formation of strong habits which ensure continued attendance. For example, Kaushal and Rhodes (2015) reported that new gym members who exercised regularly for six weeks in a consistent (i.e., stable) context were more likely to form a strong exercise habit (as assessed by the Self-Report Behavioural Automaticity Index; Gardner, Abraham, Lally, & de Bruijn, 2012). Kaushal and Rhodes (2015) further found that those who developed a strong exercise habit also reported more positive affective judgements about exercise, exercising in a more positive (i.e., supportive) environment, and engaging in exercise that was easy to do and required little effort. Encouraging strong habits may be crucial for maintaining behaviour given the typically strong correlation found between measures of habit and PA (Rebar et al, 2016).
The current study has a number of strengths. In particular, the study comprised the analysis of the attendance patterns of a large sample of all new members of a health and itness organisation, rather than a subset of members recruited into a research study. As a result, the sample should be free of any participation biases and the results more generalizable to other health and itness venues. In addition, the study employed objective measures of the frequency and stability of attendance behaviour that was recorded electronically through the member's card or wristband at the venue turnstile. In contrast, previous research focusing on the frequency and stability of exercise behaviour has relied on selfreport data (e.g., Kaushal & Rhodes, 2015) which might be open to presentational and consistency biases.
However, there are a number of study limitations that should also be noted. First, attendance data was calculated from the time members entered a health and itness venue. Although this provides objective and accurate data detailing the location, date and time of the attendance, the data output does not detail how long each member spent at the venue. Assessing the length of time members spend at a venue could help to explore the amount of PA members undertake instead of merely the day and time they attended. Second, the attendance data did not include a record of the type of activity members undertook. This would have provided some information on the likely intensity of PA members were undertaking. Both the amount and intensity of PA is likely to impact on health outcomes. Understanding the type of activity undertaken could also have been used as an additional marker of context stability to assess whether members attended the same activity, in addition to attending the same venue on the same day and the same time. Additionally, it could be that certain activities, such as exercise classes which have a predeined time, are more likely to be associated with consistent attendance behaviour. Third, the study only considered a limited range of variables that were available in the dataset. There may be other variables that are important for attendance such as goals, motivations or beliefs, which are factors identiied in models of health behaviour (Conner & Norman, 2005). In addition, factors such as the length of contract bought or the price of the membership may also inluence members' attendance behaviour. Future research could therefore assess the impact of non-renewal or shorter-term contracts and different price points on the maintenance of attendance. Fourth, it was not possible to assess any activities members were undertaking outside of the health and itness venue. It could be that some members decided to switch to undertaking PA outside of the venues during the course of their membership. Fifth, it was not possible to assess whether the members had any prior experience of using health and itness venues and the extent to which this has an impact on the maintenance of attendance. For example, some members may have already have been frequent attenders at other venues, but had taken out a new membership (e.g., due to moving to a new city). Finally, the study was conducted within a single health and itness organisation with six venues based in a large city in England. Further studies in other health and itness organisations across, and beyond, the UK are needed to establish the generalisability of these indings in a range of organisations with different facilities and attendees.
Notwithstanding these limitations, the current indings have a number of implications for policy makers and organisations providing health and itness facilities. Most importantly, the current indings highlight that attendance in new members declines over time. In particular, attendance dropped from a mean of 7.48 times in month one to 0.92 times in month 12. In addition, whereas all new members attended at least once in month one, only 22% did so in month 12. Health and itness organisations therefore face a large challenge to maintain attendance levels in new members, particularly among those who join in the winter or summer. The current indings suggest that interventions should encourage more frequent attendance in the early part of an individual's membership. Thus, health and itness venues should also seek to provide members with a positive experience in the early part of their membership to ensure that they want to return to a venue. However, it is also important for health and itness venues to put in place mechanisms that not only encourage increased attendance, but establish habitual behaviour in new members by encouraging them to attend the same venue at the same day and time each week. One way through which this could be achieved is through instructing members to form action plans detailing which venue and time they will attend (as well as which activity they will engage in). Correlational (Fleig et al., 2013) and experimental evidence (Orbell & Verplanken, 2010) indicates that engaging in action planning leads to the formation of stronger habits. Encouragingly, a recent trial has shown that an 8-week habit formation intervention resulted in a signiicant increase in PA in a sample of new gym members (Kaushal, Rhodes, Spence, & Meldrum, 2017). The intervention focused on the importance of preparation cues (e.g., preparing appropriate clothes for PA) and practice consistency (e. g., establishing a particular time for PA and stabilising the preceding events that lead to the preparatory stage of PA) as well as the use of actions plans to specify when they would engage in PA.

Conclusions
The current study examined levels of attendances and correlates of attendance behaviour in new members at UK based health and itness venues over a 12-month period. New members' early attendance behaviour, speciically the frequency and consistency of attendance, was found to be important for supporting continued attendance at health and itness venues. The indings identify a need to develop effective interventions with the potential to increase sustained attendance levels at health and itness venues.

Ethics approval and consent to participate
The study's protocol was approved by the University of Shefield's Research Ethics Committee in the School of Health of Related Research before the start of data collection (application number: 018874). Shefield City Trust provided written approval to use their data for research purposes.

Consent for publication
Not applicable.