Details of the definition of small Chinese cities are given elsewhere [15]. In short, according to the most recent Chinese census, the number of inhabitants in each of these two cities is around 5 million, and when Chinese cities are categorised via their urban population, 42.4% of the population of the entire country is shown to live in cities of 4 to 8 million [40, 41]. The vast majority of the Chinese people are living in small and medium-sized cities. Based on the Chinese city ranking carried out by China Business Network Co. Ltd, the selected cities (Yuncheng and Suihua) belong to Tiers 4 and 5 out of five tiers [42]. This means that they are both small and are of the type that is commonplace and ordinary in China. Both cities are the prefecture-level municipality and have only one central district – in Yuncheng this is Yanhu district and in Suihua, Beilin district.
The questionnaire surveys were conducted in the streets of the cities’ two central districts, in August 2017 (Yanhu) and June 2018 (Beilin), as people generally spend more time outdoors during warmer weather. Seven volunteers in Yuncheng were recruited from Yuncheng University and six volunteers in Suihua were recruited from the local urban planning department for the interviewer-administered approach. Volunteers attempted to ask every person on the city centre streets, although many busy pedestrians refused to participate in this survey. This research was approved by the ECA Ethics Committee at the University of Edinburgh (No. 06032017).
The questionnaire content with detailed items has been presented in another study [15], adapted from the Neighbourhood Environment Walkability Scale [43]. We extracted three key subscales in accordance with the aim of this study, including perceived daily walking duration as the single outcome variable (i.e. DailyWalk) and seven types of food outlet (i.e. fruit/vegetable market, fruit/vegetable street vending, snack/breakfast street vending, convenience/small grocery store, supermarket, restaurant and café/tea house). The third subscale included a range of socioeconomic characteristics: gender, age, educational attainment (junior college or lower and bachelor’s degree or higher), family income (3,000 Yuan or less, 3,001–5,000 and 5,001+) and occupation (employed, self-employed and other). During the pilot study in January 2017, food outlets used in western studies [21, 43] were adapted to the local contexts, such as street vendors and café/tea houses. A few studies in China noted that street vending could be an interesting factor to examine walkability [44–46].
We aimed to collect 200 questionnaires in each city, because the sample size of most previous studies on environment walkability was between 101 and 300 [47]. All data were carefully checked via a data-cleaning process [48]. The original datasets had 183 participants in Yuncheng and 195 in Suihua. After removing participants with missing walking duration (n = 2), missing age group (n = 1), missing education (n = 2) and missing family income (n = 1), and those aged under 18/over 60 (n = 18), the final datasets for analysis had 171 participants in Yuncheng and 183 in Suihua. Because of the small non-representative samples, we combined the two datasets (n = 354) and the city became an additional variable, which enabled regressions to be performed more feasibly.
First, we categorised DailyWalk into two levels – greater than 60 minutes and equal to or less than 60 minutes – due to the recommended daily PA being at least 60 minutes [12, 49]. The walking distances to each amenity were categorised as 1–5 minutes, 6–10 minutes, over 10 minutes and missing. “Missing” indicates “I do not know where the amenity is”. Stata 15 software was used to analyse the pooled dataset. We performed univariate and multivariate logistic regression to detect the associations between DailyWalk and the seven food outlet factors respectively. For the multivariate logistic regression model, we adjusted for city (Yuncheng and Suihua), gender, age, educational attainment, family income and occupation.
Secondly, to further understand the impact of food outlets on DailyWalk among different age groups, participants were categorised into younger adults (aged 18–35) and older adults (aged 36–59). This threshold was the same as that used in previous studies in Chinese contexts [15, 50]. We then performed the multivariate logistic regression model for the two age groups. Finally, using logistic regression, we examined the associations of DailyWalk with three levels of food environment diversity, in line with the analysis approach[8]. We set three P value thresholds of P ≤ 0.001, P ≤ 0.01 and P ≤ 0.05 for the detector sensitivity. Additionally, linear regression was performed as a sensitivity analysis by using the average walking duration on each scale to avoid bias introduced by information loss.