Built environment impact on the per capita cycling frequency of family--Based on two-level hierarchical linear model.

At present, there is less attention paid to the relationship between the frequency of travel and built environment, especially in households. In this paper, some of the determining factors in the frequency of daily cycling per household were explored based on the data from 2018 Daily Trip Survey in Xianyang, China. Then a two-level linear model was construct to identify the determining factors in the frequency of per capita daily cycling of household. According to the research results, 22.8% of the differences in the per capita cycling frequency of household are due to the differences between communities. In terms of community factors, the densities of road networks and educational facilities delivered a significantly positive impact on the per capita daily cycling frequency of family; on the contrary, the densities of medical facilities, intersections and POI delivered a significantly negative impact. Per capita cycling frequency varies considerably between households. For instance, the number of bicycles owned and the number of school-age children have a significantly positive impact on the per capita daily cycling frequency of family. However, car ownership, household income and occupation composition impose a significantly negative impact. The findings of this study would benefit the transportation engineers and planners who are keen to boost the use of active means of transportation for residents.

1. Introduction

As represents a comprehensive index that measures the travel demand of residents, travel ability and the level of urban traffic service, travel intensity can be assessed by a series of indicators, among which the number of trips per capita is most commonly used [1]. The number of trips mainly reflects the travel ability and demand of residents, while the average travel frequency per capita multiplied by the number of urban population is defined as the total trips, which is regarded as a basic measure of the minimum bearing capacity of urban transportation system.

Urban residents’ daily trips per capita is related to their social- economic development levels, population structures, city sizes, travel modes, travel time and built environments. In addition, even the same factors may have different effects on built environments, such as household income, car (bicycle) ownership, family members, family structures, land value and population density. The first four factors are often used to explore traffic trips of family members, while land value and population density are generally used to identify the trips of traffic zones. [2].

Trip generation forecast is the first stage of the four-step traffic demand forecast method. There are many approaches for trip generation forecast, of which the most commonly used ones are category generation model, growth rate method, regression generation model, etc. [3, 4]. Miller J S et al. used four different methods to calculate the differences in trip generation rates among traffic zones in nine suburban areas, and revealed that the results acquired from the direct survey, the household survey and the trip generation rate published by the American Society of Transportation Engineers would yield quite similar results, but the traffic generation rate estimated by the regional traffic demand prediction model is significantly different from the above three methods, proving that the regional applicability of the trip generation rate should be considered [5]. Zou et al. conducted a regression model to establish the relationship between population, per capita gross domestic product and travel frequency from the overall perspective of cities [6]. On this basis, the number of trips per capita in big cities and various socioeconomic indicators were analyzed by clustering to identify four relevant indicators, including urban land area, total retail sales of consumer goods, total consumer price index and per capita disposable personal income [7]. Through a regression analysis conducted in small cities, Gao et al. proposed that the influence of travel frequency indicators can be reflected in population, area, disposable personal income and per capita GDP [8]. Yuan et al. proposed a traffic generation prediction model based on the geographical advantage of the new urban district. Furthermore, Xi’an high tech Zone was exemplified to demonstrate that the method is effective in improving the accuracy of forecasting for the travel demand [9]. However, the above model often relies on the aggregate prediction method to forecast trip generation and attraction in each traffic zone. Therefore it is difficult to reflect the impact of the socioeconomic attributes of individuals and families on the travel frequency [10].

To explore travel frequencies at individual and family levels, many scholars construct traditional regression models with individual factors, family factors and built environment factors as independent variables. Stacey has studied influence factors of travel chains [11]. Parviza et al. have explored how to use logistic models to analyze complex travel chains in Hawaii [12]. Abdelghany has established two traffic assignment models based on travel chains [13]. On the basis of the above researches, al-Jammal and Golob have constructed travel chains from the perspective of individuals and families, respectively, while simulating residents’ daily travels [14, 15]. After studying the relationship between the elderly’s travel frequency and the built environment, Hanibuchi et al. revealed that population densities, parks and other facilities had significantly positive impact on this frequency [16]. Agyemang-Duah K et al. found that families’ geographical locations affect their weekday travels and shopping travels in Toronto, Canada, and then selected five traffic zones to establish an orderly response model, finally showing that after controlling the variables of family socioeconomic attributes, families’ geographical locations affect workday travel intensities and shopping travels, with the travel intensity of suburban families slightly higher than that of downtown ones [17]. Fu et al. adopted the non-aggregate model to analyze the expected value of individual trips from the perspective of an individual [18]. Based on the trip chain theory, He et al. classified travelers into three categories to construct not only the logistic regression model for student trip, but also the Poisson distribution model for elderly trips and NL model of trip choice for employment respectively. Moreover, the model was simulated to get trips of different categories of travelers through combination with the macro social and economic data. [19]. The default assumption in above models is that the micro-individual space is irrelevant, which ignore that each individual is also a member of a family or a community group and his/her travel behavior is also affected by the structural characteristics of his/her family and community at different levels. Therefore, it is impossible to accurately reveal the intrinsic association between built environments and travel behaviors.

In addition, with the popularization of big data, machine learning algorithms are also widely used in the prediction of travel volume, and the neural network model is a common method. Chen et al. introduced the Long Short-Term neural network to fulfill the traffic flow prediction, the results show that the traffic flow prediction accuracy of the hybrid neural network model combined with denoising method is better than that of the non-hybrid neural network model [20]. then, the data denoising scheme is combined with the deep learning model, and the EMD, EEMD and WL models of different bases are introduced to identify and correct the noise traffic flow data samples, experimental results show that the hybrid traffic flow prediction scheme is better than the prediction process [21]. However, such methods have problems such as large amount of sample data, long training time and difficult to explain.

To sum up, most of the previous studies take the traffic zone or individual as the research unit and lack of attention to family as the research unit. Furthermore, motorization-oriented generative prediction modeling has focused attention on cars, and the scale of these models is not the same as that of cycling, besides, the behaviors of groups from the same spatial are similar but there are differences between groups in different spatial regions. The existing methods ignore the intra-group homogeneity and inter-group heterogeneity of such nested data, which can lead to biased conclusions. In addition, previous studies paid more attention to the built environment and population density, traffic facility density, POI diversity and other conventional indicators. The safety of riding facilities and the influence of different types of facility density on riding facilities need to be further studied.

In order to overcome the limitations of most of the previous studies, this study construct a two-level linear model based on data from 2018 Daily Trip Survey in Xianyang, China and select the family attributes and built environment as dependent variables, to explore the influence of spatial heterogeneity on travel frequency. This study not only provides a comprehensive overview of factors that influence per capita cycling frequency of family, but also quantifies the spatial heterogeneity of the cycling frequency and reveal the role of communities’ built environments in families’ cycling frequency. The results of the study can provide technical support for urban planning, policy designing, boosting green urban travels, reducing transportation energy consumption and carbon emissions, and make cities healthier and more sustainable places to live and work.

This paper is organized as follows. The next section illustrates the data sources and the influencing factors, as well as indicators to measure built environment. Section 3 introduce the research methodology and the zero model, random intercept model and full model are constructed. Section 4 analyzes the model results, and the effect of different level factors are further explicated. The results are discussed in the last section.

2. Material and research methodology

2.1 Study area and data sources

As part of Shaanxi Province, Xianyang city is located in northwest China and adjacent to the provincial capital Xi ’an in the east. Covering a total of 10196 square kilometers, it has a permanent population of 395,9842 and an annual GDP of CNY 220.481 billion. Taken as research object for this study, the relatively well-developed city center of Xianyang covers an area of 60 square kilometers, including 760000 people. The travel data was obtained through a two-stage stratified sampling survey. In the first one, the whole sampling method was applied to select the communities to be surveyed. In the second and last one, the survey households were randomly selected from each community. Besides, all-day travel survey of family members aged over 6 years was conducted in the form of home visits. It should be noted that, in the first page of the questionnaire, we introduced our research purpose and secured their written informed consent to participate in this academic research. Non-adult participants were allowed to answer this questionnaire only after getting their parent or guardian’s written informed consent. The Chang’an University Ethics Committee approved the protocol and informed consent forms for this study.

The household survey was conducted during the period ranging from May 29 to May 31, 2018, with a total of 5,080 households and 13,607 respondents investigated and a success rate of 100% reached. The dependent variable in this study is the per capita cycling frequency of family. 108 communities and 2,934 families were found meeting the requirements of this research, and the minimum sample of families in each community was not less than 10. According to Goldstein, the sample size should be no less than 10 for each group and no less than 100 for level 2 to reduce sampling errors and balance the sample size between groups [22]. Given that there were 108 communities in level 2, each with a samples of 10 or more families, the sample size meets the requirements of subsequent model analyses. The community boundaries as demonstrated for the research units are shown in S1 Fig.

2.2 Variable selection and description

2.2.1 Family socio-economic attributes

According to previous studies, family structure, family income, household car and bicycle ownership are all associated with travel behavior [23-26]. The household socioeconomic attributes include total household population(PEO), household occupational composition(PRO), monthly household income(INCOM), the number of school-age children(CHIL), the number of elderly(OLD), ownership of vehicles and other attributes may affect travel behavior. Variables description is shown in Table 1.

Table 1

Descriptive statistics for family socio-economic attributes (N = 2934).

Variable	Question options	n	Percentage (%)	Variable	Question options	n	Percentage (%)
PEO	1	110	3.7	OLD	0	2448	83.4
	2	1082	36.9		1	186	6.3
	3	1443	49.2		2	300	10.3
	4	234	8.0	CHIL	0	1519	51.8
	5	58	2.0		1	633	21.6
	6	7	0.2		2	649	22.1
BIKE	0	1217	41.5		3	133	4.5
	1	1313	44.8	INCOM	<CNY3000	513	17.5
	2	379	12.9		CNY3001-10000-	2290	78.0
	3	25	0.8		>CNY10000	131	4.5
CAR	0	1320	45.0	PRO	double workers	1774	60.5
	1	1555	53.0		single workers	760	25.9
	2	58	2.0		jobless	400	13.6
	3	1	0.0

Note: CHIL refers to primary, middle and high school students aged between 6 and 18.

As can be seen from Table 1, the majority of the families have 2–3 members, mainly with a double-worker composition, followed by single-worker. And jobless families take the lowest proportion. The largest part of the families report a monthly income of CNY3001–6000, representing 78%, followed by below CNY3000, representing 17.5%, high income families (with monthly income of above CNY10000) account for 4.5%. For the ownership of transportation, 55% of households have at least one car, and 58.2% of households have at least one bicycle, families with two school-age children and one school-age child had almost the equal proportion.

2.2.2 Built environment

In 1997, Cervero and Kockelman proposed a "3D" (density, diversity and design) model for the measurement of built environment [27, 28]. Then, Cervero et al. introduced accessibility and distance into 3D elements and transformed them into a "5D" model for studying Bogota built environment and its impact on walking and cycling behavior. Other scholars used “5D”model to study the correlation between travel characteristics and built environment in subsequent studies [29, 30]. Combination with the availability of data, this study divides the built environment variables into three categories: community population density, road traffic infrastructure and other daily life service facilities, including 12 built environment indicators.

Community population density

Population density is regarded as an important influencing factor towards travel intensity. An empirical study on the influence of urban built environment on commuting mode choice in Shanghai shows that the population density is positively correlated with the probability of non-motor vehicle use [31]. According to a study in Maryland, the higher the population density is, the higher the possibility of carrying bicycle is, In place with higher industrial employment, the use of bicycles increase relatively [32]. A study on the relationship between built environment and physical activity shows that the population density, parks or green spaces are positively correlated with activity frequency regardless of the buffer zone chosen [16]. In this study, the population density was calculated by the total population in a community /community area.

Road transportation infrastructure

In the most relevant studies, the road infrastructure indicators mainly included road network density, the road section, intersection density, bus station density, subway station density, etc. [33, 34]. This study used the road network density, intersection density and bus station density as the indicators of road traffic infrastructure, which were calculated as follows:

Other daily life service facilities

Many studies take land-use density and land-use diversity as quantitative indicators of other service facilities in the built environment and obtain consistent conclusions by analyzing the relationship between them and travel behavior [35-37]. A few studies used POI as an indicator of built environment [38, 39]. This study took POI density, POI diversity, and the POI density of four main daily life facilities (including school, shopping, leisure, and medical facilities) as the indicators of other daily life service facilities, as follows:

Based on the existing literature [40], the community POI diversity is defined as follows:

Where P represents the probability, A indicates the number of type j POI facilities in the i-th community, and A denotes the total number of POIs in the i-th community. In the formula, H refers to the spatial information entropy. It can be find out that when H≥0, the level of spatial information entropy can be used to indicate the mixing degree of various facilities in the community. That is to say, higher entropy indicates more types of facilities.

The following figure (S2–S10 Figs) showed the attribute information about population density(S2), POI diversity(S3), POI density(S4), road network density(S5), intersection density(S6), bus station density(S7), the POI density of education, shopping, leisure, and the medical services(S8), the proportion of motor and non-motor vehicle separate railing (S9) and the proportion of non-motor lane parking ratio(S10) in each community.

As can be seen from S2, the population density gradually decreases from the center to the periphery, and the population density in the areas with dense commercial service facilities is higher than that in the new urban areas. It can be seen from S3 and S4 that both the POI diversity and POI density show a typical pattern of contiguous distribution and spread along the line. Contiguous distribution mainly includes the old street area, the business circles on Maotiao Road, the business circles of 505 Square an others, while the spread line mainly includes Renmin Road, Yuquan Road, Weiyang Road, etc. As for the new urban areas, both the diversity and density of POI are low.

Seen from S5, S6, S7 and S8, the densities of road network, intersection, bus station and public service facility around the old urban areas and commercial core areas are higher than those of the urban fringe areas and newly built areas. S8 and S9 show that the traffic volume of the commercial core areas is large and most of the roads are equipped with motor and non-motor vehicle separate railings. The roads of the new areas in the west and south were built under higher construction standards with cross section of three or four boards roads in the main and secondary roads, therefore, the motor and non-motor vehicle separate railings account for a relatively high proportion. From S10, it can be seen that the proportion of non-motor lane parking in the western part of the built-up area is more serious than that in the eastern part. For one thing, there are relatively few traffic monitoring facilities in the west and the law enforcement is weaker than that in the east. For another, there are fewer public parking lots in the west areas compared with the east areas.

2.2.3 Per capita cycling frequency of family(PCF)

The existing studies usually took different periods of travel time (week, day, peak hour or hour), arrival/departure rates and travel frequency as dependent variables [41, 42]. However, this research mainly studies the influence of community built environment on family cycling frequency. The relationship between cycling frequency and built environment is more significant than that between cycling time and built environment [43]. Therefore, take the per capita cycling frequency of family(PCF) as dependent variable, which was calculated by dividing the total daily cycling frequency of family members by the number of family members over 6 years old.

2.3 Methodology

In order to study the effect of family social-economic characteristics and built environment on the per capita cycling frequency of family, a two-level hierarchical linear model (HLM) was applied. Due to the nested data structure of family travel frequency within community, HLM was considered as the most appropriate method that can be used to analyze this data. All analyses were conducted using HLM7 software. The Level-1 variables included the total household population, household income, the number of children, the number of the old and the occupation of family member. The Level-2 variables included the density of community population, the density of road network, intersection density, the density of bus station, POI diversity, POI density and the density of education, medical, shopping and leisure facilities, all of which were closely related to the daily life of residents. The community ID corresponding to each family represented the unique connection point between the variables of level-1 and level-2.

Modeling a multi-layer model is a step-by-step process. Combining the methods recommended by Hox (1994) and Singer (1998), the construction process consists of the following three mode is: null mode, random intercept model and full mode. The null model can be used to assess within-group or between-group homogeneity and only when there is significant intra-group correlations between them, then a multi-level linear model can be established. The random intercept model includes random effect regression models as well as regression models that use means as outcome measures. This model can explore the influence of household level variables and regional level variables on the per capita cycling frequency of family. The full model is used to explore whether the level-2 variables deliver a significant impact on the intercept and coefficient of the level-1 model.

2.3.1 Null model

Null model or intercept-only model, which is identical to a one-way random effect analysis of variance, the general form of null model is as follows.

While, In level-1 Y0 indicates the per capita cycling frequency of family(PCF), β0 represent mean outcome of group j, r represent the random individual variable around this mean. γ00 denotes the overall intercept representing the grand mean of Y, μ0 captures the variable between group means.

2.3.2 Random intercept model

Random intercept models include random effects regression models and regression models with the mean as the outcome variable. Thereinto, the random intercept regression model is used to study the level-1 variables, that is, the effect of family level factors on cycling frequency; the random effects regression model is used to study the level-2 variables, that is, the effect of community level factors on cycling frequency.

Random effects regression model

The random effects regression model only involved the family level variables and was used to test the correlation between the level-1 variables and the outcome variables. The random effects regression model was established as follows:

While, β0 in level-1 represents the average result of the j-th unit, r denotes the error of each item in level-1, β1, β2, β3, β4, β5, β6, β7, β8 and β9, respectively indicate the total household impact coefficients of population, family car ownership, family bicycle ownership, number of school-age children in the family, number of elderly in the family, low-income families, high-income families, single-employee families, and workless families. γ00 in level-2 refers to the overall mean of the results, and μ0 refers to the random effect associated with the j-th unit.

Regression model with the mean as the result

In order to further analyzed the impact of community level factors on the per capita cycling frequency of family, this paper used the average of the level-2 variables as the estimation result to establish a regression model, which was as follows:

While, β0 in level-1 represents the average result of the j-th unit, and r denotes the error of each item in level-1.

γ00 in level-2 refers to the overall mean value of the result, γ00, γ01, γ02, γ03, γ04, γ05, γ06, γ07, γ08, γ09, γ10, γ11 and γ12 respectively indicate community population density, community POI diversity, and community POI density, road network density, intersection density, bus stop density, POI density of education, shopping, leisure and medical, proportion of non-isolation bar, and proportion of non-motor vehicle parking. μ0 refers to the conditional variance of β0 after the control variable.

2.3.3 Full model

The full model was used to study the average intercept of the level-1 model and the variation of coefficients at the level-2. The level-1 and level-2 variables of random intercept model are brought into the full model for modeling. The specific model was as follows:

Where, β0 in level-1 represents the average result of the j-th unit, β1, β2, β3, β4, β5, β6 and β7 respectively indicate the total family population, family car ownership, family bicycle ownership, family school-age children, number of elderly in the family, low-income households, high-income households, single-employee households, and workless households. r denotes the error of each item in level-1. γ00 in level-2 refers the average intercept of all units in level-2. γ00, γ01, γ02, γ03, γ04, γ05 respectively indicate the community POI diversity, road network density, intersection density, POI density of education and medical on the per capita cycling frequency of household, γ00, γ10, γ20, γ30, γ40, γ50, γ60, γ70 refer the average regression slopes of all units in level-2, γ20 and γ31 indicate the influence coefficients of road network density on the bicycle ownership and school-age children riding in level-1, γ41, γ51, γ61 and γ71 indicate the influence coefficient of community POI diversity on the cycling frequency of high and low-income families in level-1, and γ42 and γ52 denote the impact coefficient of road network density on the cycling frequency of high- and low-income families in level-1. γ62 and γ72 denote the influence coefficients of educational facilities on the cycling of single-employees and unemployed families on level-1.

3. Results

3.1 Null model estimation

To begin with, an unconditional model was tested to determine whether there were significant variation existing at each level of the hierarchy. The parameter estimation were shown in Table 2, from which can be seen that there was a significant differences between groups(). That was to say, there were significant inexplicable differences among community.

Table 2

Parameter estimation of null model on the per capita cycling frequency of family.

Random Effect	Standard Deviation	Variance Component	Df	Chi-square	P-value
U0	0.577	0.33	107	947.11	0.000
R	1.057	1.12

Note: ICC = 0.228.

According to Cohen, an ICC is lower than 0.059 which indicates a weak intra-group correlation, the range from 0.059 to 0.138 suggests a moderate correlation, and an ICC higher than 0.138 which indicates a strong correlation. In general, as long as ICC(1) exceeds 0.059, it is necessary to consider inter-group effects in statistical modeling processing [44]. The intra-group correlation coefficient (ICC) of the per capita cycling frequency of family was 0.228, which means that about 22.8% of the total variation in the outcome measure was attributed to the variations between communities, constituting a significant intra-group correlation. In this case, it was not suitable to adopt the general regression model and a multilevel model was required for data analysis.

3.2 Random intercept model estimation

3.2.1 Random effects regression model estimation

Based on null model, a random coefficient regression model was introduced, under which only the level-1 had predictor variables, while level-2 had none. The level-1 predictor variables included the total family population, the number of owned cars, the number of owned bicycles, the number of school-age children in a family, the number of elders in a family, the monthly household income, and the family occupational composition. The parameter estimation of the random coefficient regression model was shown in Table 3.

Table 3

The parameter estimation of the random coefficient regression model.

Fixed effect	Coefficient	Standard error	T-ratio	d.f.	P-value
INTRCPT1	1.431	0.130	10.999	107	0.000
PEO	0.056	0.040	1.423	2924	0.152
CAR	-0.359	0.048	-7.467	2924	0.000
BIKE	0.367	0.040	9.151	2924	0.000
CHIL	0.080	0.029	2.790	2924	0.006
OLD	0.011	0.047	0.226	2924	0.821
INCOM1	0.239	0.071	3.378	2924	0.001
INCOM3	-0.183	0.072	-2.557	2924	0.011
PRO2	0.150	0.051	2.939	2924	0.004
PRO3	0.301	0.101	2.963	2924	0.004

Note: The parameter estimates was based on robust standard error

The results demonstrated that the variables included the total household population and the number of elderly in level-1 had no significant impact on the per capita cycling frequency of family, while other variables included car ownership, bicycle ownership, number of school-age children in a family, family income and family occupation composition had statistically significant impact on the per capita cycling frequency of family.

It can be seen from Table 3 that in the level-1 predictor variables, car ownership had a significant negative impact on the per capita cycling frequency of family(sig. = 0.000<0.001), That was to say, the car ownership increases by one unit, the per capita cycling frequency of family was reduced by 0.359 units. while, bicycle ownership had a significant positive impact on the per capita cycling frequency of family (sig. = 0.000<0.001), suggesting that the number of family bicycles increased by one unit, the per capita cycling frequency of family increased by 0.367 unit; the number of school-age children in the family had a significant impact on the frequency of family cycling (sig. = 0.006<0.01, regression coefficient is 0.080). There was a significant difference in the per capita cycling frequency of family between low-income, high-income families and middle-income families. (Sig. = 0.001<0.01 and sig. = 0.011<0.05), among them, the per capita cycling frequency of low-income families was 0.239 units higher than that of middle-income families and the per capita cycling frequency of high-income families was 0.183 units lower; single-employee families were significant differences in the per capita cycling frequency between workless families and dual-employee families (sig. = 0.004<0.01 and sig. = 0.004<0.01), and the per capita cycling frequency of single-employee families and workless families were 0.150 and 0.301 units higher than that of dual-employee families, respectively.

3.2.2 Regression model with the mean as the result estimation

On the basis of the zero model estimation, the level-2 had predictor variables and the level-1 had none. The level-2 predictive variables included community population density, community POI diversity, community POI density, road network density, intersection density, bus station density, POI density of education, shopping, leisure, medical, the proportion of non-isolation bar and the proportion of non-motor lane parking ratio. The parameter estimation of the regression model based on the mean value was shown in Table 4.

Table 4

The parameter estimation of the regression model based on the mean value.

Fixed Effect	Coefficient	Standard error	T-ratio	d.f.	P-value
INTRCPT1	1.832	0.055	33.368	95	0.000
PCPEO	-0.000	0.000	-0.678	95	0.499
POID	-0.823	0.262	-3.145	95	0.003
PPOI	0.121	0.081	1.491	95	0.139
PNET	0.003	0.001	2.735	95	0.008
PINT	-0.003	0.001	-2.143	95	0.034
PBUS	0.001	0.003	0.217	95	0.829
PEDU	0.021	0.005	4.315	95	0.000
PSHO	0.007	0.045	0.152	95	0.880
PLEI	-0.001	0.004	-0.197	95	0.844
PMED	-0.002	0.001	-2.761	95	0.007
PPOK	0.129	0.090	1.430	95	0.156
PISO	0.006	0.009	0.655	95	0.514

Note: The parameter estimates in the table are based on robust standard error

The results showed that in the level-2 predictor variables included the community population density, POI density, bus station density, shopping facilities, leisure facilities, the proportion of non-isolation bar and the proportion of non-motor lane parking ratio had no significant impact on the per capita cycling frequency of family, while other level-2 predictive variables included POI diversity, road network density, intersection density, educational facilities and medical facility density had reached Statistically significant level.

It can be seen from Table 4 that in the level-2 predictor variables, POI diversity, intersection density, and medical facilities had a significant negative impact on the per capita cycling frequency of family(sig. = 0.003<0.01, sig. = 0.034<0.05 And sig. = 0.007<0.01), that was, for each additional unit in POI diversity, intersection density, and medical facility density, the per capita cycling frequency of household would be decrease by 0.823, 0.003, and 0.002 units respectively, the road network density and the density of education facilities had a significant positive impact (sig. = 0.008<0.01 and sig. = 0.000<0.001), that was, for each additional in road network density and educational facility density, the per capita cycling frequency of family would be increased by 0.003 and 0.021 unit respectively.

3.3 Full model estimation

According to the results of the random coefficient regression model, the variables in the level-1 that had no significant impact on the per capita cycling frequency of families were eliminated, while those with significant influences variables were retained. Therefore, the predictor variables that were adopted in the full model were: car ownership, bicycle ownership, the number of school-age children in a family, family income, and family occupational composition. Similarly, predictive variables in the level-2, including POI diversity, road network density, intersection density, education facility density and medical facility density were kept in the full model.

The parameter estimates of the full model are shown in Table 5, which suggested that the diversity of POI and road network on the community level had no significant impact on the per capita cycling frequency of family whereas the density of intersections, educational and medical facilities exerted a significant effect on the per capita cycling frequency of family. Besides, the density of educational facilities exhibited a significant positive impact on the per capita cycling frequency of family as controlling for other relevant variables(p = 0.000<0.001). That is to say, for each unit increase in the density of educational facilities, the per capita cycling frequency of family would be increased by 0.015 unit, suggesting that educational facilities can be used to partially explain the differences in the per capita cycling for families in the community. Moreover, the density of intersection and medical facility had a significant impact on the per capita cycling frequency of family(p = 0.000<0.001 and P = 0.009<0.01). Contrary to what was expected, the density of intersection and medical facility had a significant impact on the per capita cycling frequency of family. One possible cause for this is that the high density of intersections may lead to frequent cycling interruptions. And furthermore, the medical facilities were usually used in case of physical discomfort, therefore, most of the trips relay on motor vehicles.

Table 5

The parameter estimates of the full model.

Fixed Effect	Coefficient	Standard error	T-ratio	d.f.	P-value
INTRCPT	1.547	0.087	17.712	102	0.000
POID	-0.281	0.199	-1.413	102	0.161
PNET	-0.001	0.002	0.710	102	0.479
PINT	-0.004	0.001	-4.104	102	0.000
PEDU	0.015	0.004	4.100	102	0.000
PMED	-0.001	0.001	-2.677	102	0.009
CAR	-0.346	0.048	-7.246	2911	0.000
BIKE	0.370	0.039	9.548	2911	0.000
BIKE*PNET	0.001	0.000	2.621	2911	0.009
CHIL	0.119	0.026	4.479	2911	0.000
CHIL*PNET	0.002	0.001	2.127	2911	0.033
INCOM1	0.213	0.068	3.139	2911	0.002
INCOM1*POID	-0.368	0.370	-0.994	2911	0.321
INCOM1*PNET	-0.002	0.002	-1.139	2911	0.255
INCOM3	-0.163	0.065	-2.524	2911	0.012
INCOM3*POID	-0.455	0.099	-4.577	2911	0.000
INCOM3*PNET	0.004	0.003	1.289	2911	0.198
PRO2	0.147	0.050	2.971	2911	0.003
PRO2*POID	0.160	0.169	0.946	2911	0.345
PRO2*PEDU	0.008	0.005	1.726	2911	0.084
PRO3	0.309	0.080	3.878	2911	0.000
PRO3*POID	0.460	0.213	2.159	2911	0.031
PRO3*PEDU	0.008	0.002	3.680	2911	0.000

Note: The parameter estimates in the table are based on robust standard error

Among the level-1 predictor variables, the household car ownership still had a significantly negative impact on the per capita cycling frequency of family (sig. = 0.000<0.001) with a regression coefficient of -0.346, indicating that the per capita cycling frequency of family would be reduced by 0.346 units with per one unit increase in household car ownership. The household bicycle ownership and the number of school-age children in a family still had a significantly positive impact on the per capita cycling frequency of family (sig. = 0.000<0.001 and sig. = 0.000< 0.001) with regression coefficients of 0.370 and 0.119, respectively. The more the bicycles and school-age children in a family had, the higher the per capita cycling frequency of family was. The frequency of low-income and high-income families was significantly different from that of middle-income ones (sig. = 0.002<0.01 and sig. = 0.012<0.05). In addition, the per capita cycling frequency of low-income families was 0.213 units higher than that of middle-income ones, while the per capita cycling frequency of high-income families was 0.163 units lower. There was a significant difference in the per capita cycling frequency between single-employee, workless and dual-employee families (sig. = 0.003<0.01 and sig. = 0.000<0.01). The per capita cycling frequency of single-employee and workless families were 0.147 and 0.309 units higher than that of dual-employee families, respectively. To sum uo, family factors delivered significantly greater impact on a family’s per capita cycling than community factors.

On cross-level interaction, the density of road network promoted cycling behavior among those families with bicycles and school-age children. Statistically, the cross-level interaction coefficients between road network density and bicycle ownership, the number of school-age children were 0.001 (sig. = 0.009<0.01) and 0.002 (sig. = 0.033<0.05), while the interaction between POI diversity and high-income households had a more significant negative impact compared with middle-income households, with the interaction coefficient reaching -0.455 (sig. = 0.000< 0.001). Besides, the interaction between POI diversity and jobless families exerted a more significant positive effect compared with dual-income families, with the interaction coefficient reaching 0.460 (sig. = 0.031<0.05). In addition, the density of educational facilities showed a significant difference in promoting cycling behavior between jobless families and dual employees, with the interaction coefficient reaching 0.008 (sig. = 0.000< 0.001). This may result from the fact that jobless families tended to choose the school closer to their residence and the children from this family were more inclined to cycling. In comparison, the dual-income families have higher incomes and pay more attention to the quality of educational facilities, with the children from these families are usually driven to and from school by their parents.

4 Conclusion and discussion

Analysis based on survey data, this study employ a two-level linear model to identify the main factors affecting the per capita cycling frequency of family, which focus on the impact and interaction of this frequency at two different levels of the family and the community. The results show that 22.8% of the heterogeneity in the per capita cycling frequency of family comes from communities, which means that more differences among these families. That is to say, family factors are more important in explaining the differences in the households per capita daily cycling frequency, because each individual trip in a family are often restricted by the family social-economic attributes and the division of among family members, while the influence of communities is relatively weak. This results are consistent with the findings of international research. Surveys on Shanghai residents’ cycling reveal that the frequency of using shared bicycles are mainly affected by age, education, income and family size, etc., followed by riders’ satisfaction with the riding environment safety and comfort [19]. W. Wendel-Vos1’s research demonstrates that a few hypothetical environments have been supported as determinants of active travels such as cycling [45]. Ding studied the impact of family structures and built environments on travel behavior of the elderly and non-elderly and the results showed that whether living with children, population density and the degree of mixed land use affect the travel characteristics of the elderly [46].

At family factor aspect, the bicycle ownership and the number of school-age children in the family have a significantly positive impact on the per capita cycling frequency of family, while the car ownership, family income and family occupation composition have a significantly negative impact on the per capita cycling frequency of family. That is to say, the relevant policies such as raising the threshold of vehicle ownership can improve the frequency of cycling.

As far as community factors are concerned, the density of educational facilities has a significantly positive impact on the per capita cycling frequency of family, suggesting that measures such as enriching the education facilities in communities would help boost cycling. Contrary to what is expected, both the POI density of medical facilities and the intersection density have significantly negative impact on the per capita cycling frequency of family. One possible explanation for this phenomenon was that the high intersection density may make frequent interruptions for cycling; another possible explanation is that the medical facilities were required in case of physical discomfort, therefore, most of trips relay on motor vehicles.

Then a cross-level interactions indicate that the density of road network promote cycling in families with bicycles and school-age children. In addition, the density of educational facilities has a more significant difference in promoting cycling among the jobless families than that of dual employees, which may be attributed to the fact that jobless families tend to choose the school closer to their residence and the children from this family are more inclined to cycling. In contrast, the dual-income families have higher incomes and pay more attention to the quality of educational facilities and the children from these families are usually driven to and from school by their parents.

To sum up, diversity of public service facilities, good road infrastructure and appropriate policies are conductive to the cycling safety and comfort for community members to promote cycling. Same specific measures are as follows: firstly, on the urban construction, reasonable arrangement of education, medical and other facilities adjacent to home areas can ensure that daily travel distances of residents are within a proper range of cycling. Secondly, it is necessary to improve the safety of cycling environment by establishing bicycle lanes and other facilities. Finally, if the existing built environment is difficult to optimize, policies such as increasing the barrier to purchasing cars shall be considered, so as to curb the use of motor vehicles and promote active travel behavior.

This study also has some limitations, Firstly, the model can be further expanded into a three-level model by adding individual level factors in the future. Secondly, other factors such as residents’ personal values, group identity, weather, as well as terrain can be considered in the future research. Finally, the sample size can be further expanded and the groups can be divided with travel purposes and age stages, which will help transportation engineers and planners to put forward feasible strategy and targeted tactics to increase the use of active means of transportation for residents.

Research scope and community boundaries.

(TIF)

Click here for additional data file.

Population density.

(TIF)

Click here for additional data file.

POI diversity.

(TIF)

Click here for additional data file.

POI density.

(TIF)

Click here for additional data file.

Road network density.

(TIF)

Click here for additional data file.

Intersection density.

(TIF)

Click here for additional data file.

Bus station density.

(TIF)

Click here for additional data file.

The POI of shopping, leisure, education and medical.

(TIF)

Click here for additional data file.

The proportion of non-isolation bar.

(TIF)

Click here for additional data file.

The proportion of non-motor lane parking ratio.

(TIF)

Click here for additional data file.

(XLSX)

Click here for additional data file.

21 Feb 2022

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Quan Yuan, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: 1. The authors were recommended to ask professional editors or native speaker to polish the manuscript language due to that many sentences were difficult to follow. E.g., “The data was based on the travel survey data collected from the residents in Xianyang in 2018 and the POI data…”

2. I am not so convinced about the statement “Traffic generation and attraction forecast are the first element of the traditional four stage prediction method, which can be classified into growth rate method, original unit method (cross classification or type analysis) and function method…” Please add more explanations.

3. The authors were recommended to better summarize the academic contributions in the INTRODUCTION section.

4. The authors were recommended to merge the study area section and data collection section.

5. Please delete Chinese symbol 元 in table 1 considering that the journal is an international journal with potential readers were around the world.

6. Pleas number each equation in the manuscript.

7. The following studies were recommended to be properly cited: [1] Traffic flow prediction by an ensemble framework with data denoising and deep learning model, Physica A: Statistical Mechanics and its Applications, vol. 565, p. 125574, 2021.[2] Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison, IEEE Sensors Journal, vol. 20, pp. 14317-14328, 2020.

Reviewer #2: I find the idea has certain application value; Following are my comments after reading your manuscript.

1. It is suggested to further explain the consistency between the groups of the questionnaire and the actual situation of urban citizens.

2. Figure 2-figure 10. These nine figures are just a simple list in the main body of the paper. It is suggested to add corresponding discussion and explanation.

3. This paper constructs three models, but the construction premise of the three models is not clear, and the relationship and research value of the three models are not explained in detail.

4. The innovation and practical application scenarios of the paper need to be further elaborated.

I hope my comments may be useful for the authors.

Best wishes.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

6 Apr 2022

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Built Environment Impact on the Per Capita Cycling Frequency of family”(ID: PONE-D-21-40172). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. The changes in the revised manuscript have been highlighted by red color. Point by point responses to the reviewers’ comments are listed below this letter.

Responds to the reviewer’s comments:

TO REVIEWER 1：

Response:

Thank you very much for the detailed comments. We appreciate your time and help in reviewing our manuscript, and the insightful comments you provided that have helped significantly improve the quality of this study. We have revised the paper very carefully according to your suggestions, and detailed explanations of all the issues are as follows.

1. The authors were recommended to ask professional editors or native speaker to polish the manuscript language due to that many sentences were difficult to follow. E.g., “The data was based on the travel survey data collected from the residents in Xianyang in 2018 and the POI data…”

Response:

Many thanks for pointing out this issue.

We apologize for the poor language of our manuscript. We have now worked on both language and readability and we also leverage 3rd party service for language polishing. The English of the full manuscript has been strongly revised in the following aspects:

1. We have improved the language expression logic to guide the readers better.

2. We have reorganized language expressions for the verbose passages to make it easier for readers to understand.

3. We have revised the grammar and typos in the original text.

4. We have enriched vocabulary expressions.

5. We have tried to avoid the use of first-person pronouns.

All modified parts are marked in red font in the updated manuscript.

2. I am not so convinced about the statement “Traffic generation and attraction forecast are the first element of the traditional four stage prediction method, which can be classified into growth rate method, original unit method (cross classification or type analysis) and function method…” Please add more explanations.

Response:

Many thanks for your question. We elaborated this issue as follows:

I am sorry that this part was not clear in the original manuscript. We have reorganized this passage with reference to relevant literature as follows:

Trip generation forecast is the first stage of the four-step traffic demand forecast method. There are many approaches for trip generation forecast, of which the most commonly used ones are category generation model, growth rate method, regression generation model, etc.[1,2] However, the above model often relies on the aggregate prediction method to forecast trip generation and attraction in each traffic zone. Therefore, it is difficult to reflect the impact of the socioeconomic attributes of individuals and families on the travel frequency[3].

[1] Liu An, Song Weifang, Yang Peikun. Reaearch on travel generation prediction model. Journal of Tongji University(Natural Science). 1998.26(3):290-293.

[2] Shi Fei, Wang Wei, Lu Jian. Induction and innovation of resident trip generation prediction methods. 2005,3(1):43-46.

[3] Yang Min, Chen Xuewu, Wang Wei et al. Commuters trip generation model based on activity patterns. Journal of Southeast University( Natural Science Edition) 2008, 38 ( 3) :525-530.

The modified contents have been marked in red font on Page 3 line 48-51 and Page 4 line 69-71 in the updated manuscript.

3. The authors were recommended to better summarize the academic contributions in the INTRODUCTION section.

Response:

Thanks for your suggestion.

We have rewritten the Introduction as suggested by the reviewer. According to the different basic travel units of research, we have discussed the aggregate model constructed by traffic zones and the disaggregated model constructed by individuals or households, separately. At the same time, we have added relevant research on machine learning model, with new references of [3,4,10,20,21]. In addition, we have reorganized the academic contributions of this research from the aspects of research perspectives, research elements and research methods.

At research perspectives aspect, this research focuses on the travel generation prediction of cyclists instead of all-mode resident travel generation prediction model. It is committed to studying the correlation mechanism between the per capita cycling frequency of household, household attributes and built environment elements, which promotes green and healthy lifestyle.

As far as research elements are concerned, the built environment elements of existing research mainly include population density, traffic facility density, POI diversity and other conventional indicators. This research has not only included two variables of motor and non-motor vehicle separate railing ratio and non-motor lane parking ratio that affect cycling safety, but also included some factors that are closely related to daily life and have a significant impact on the per capita cycling frequency of household, such as the density of education, shopping, leisure and medical facilities.

From the aspect of research methods, this research has broken through the defect of losing data information in the disaggregating or aggregating process of traditional regression models. According to the nested structure of family-community data and considering the within-group homogeneity and between-group heterogeneity of the data, a multi-level linear regression model was used in this research to explore the effects of household and community level variables on the per capita cycling frequency of household and the cross-layer interaction effects between household and community.

The modified contents have been marked in red font on page 5 line 110-112 and page 6 line 113-119 in the updated manuscript.

4. The authors were recommended to merge the study area section and data collection section.

Response:

Many thanks for your suggestion.

We have merged the area section and data collection section according to the comments of the reviewer, and the chapter number was readjusted accordingly.

The modified contents have been marked in red font on Page 7 line 135-154 and Page 8 line 155-160 in the updated manuscript.

5. Please delete Chinese symbol 元 in table 1 considering that the journal is an international journal with potential readers were around the world.

Response:

Many thanks for pointing out this issue.

We apologize for the typos of our original manuscript, in the revised manuscript, we have corrected the figure and revised “元” to CNY.

The modified contents have been marked in red font on Page 8 line 169 and Page 6 line 173-175 in the updated manuscript.

6. Pleas number each equation in the manuscript.

Special thanks to you for your suggestion.

We have numbered each equation in the updated manuscript.

The modified contents have been marked in red font on Page 11 line 223-225, Page 14 line 291-292 and line 308, Page 15 line 309-318 and line 330, Page 16 line 333,347-352 and Page 17 line 352-358 in the updated manuscript.

7. The following studies were recommended to be properly cited: [1] Traffic flow prediction by an ensemble framework with data denoising and deep learning model, Physica A: Statistical Mechanics and its Applications, vol. 565, p. 125574, 2021.[2] Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison, IEEE Sensors Journal, vol. 20, pp. 14317-14328, 2020.

Special thanks to you for your good suggestion.

Reference:

The paper you provide has a great effect on a comprehensive understanding of the method of forecasting trip generation, so we added this article as reference [20,21] in the revised manuscript, and We will explore this aspect in the next work.

The modified contents have been marked in red font on page 5 line 99-107 in the updated manuscript.

TO REVIEWER 2:

Response:

We appreciate your time and help in reviewing our manuscript. Thank you very much for your affirmation of our work in this paper and many thanks for your detailed comments. We have revised the paper very carefully according to your suggestion, and all the modified contents have been marked in red font in the updated manuscript.

1. It is suggested to further explain the consistency between the groups of the questionnaire and the actual situation of urban citizens.

Response:

Many thanks for pointing out this issue.

This study based on the data from 2018 Daily Trip Survey in Xianyang, China. The total sample size is 7,070 persons of 2,934 households with an average population of 2.41, including 48.8% males and 51.2% females. As for the age structure, the main group is 15–59 years old(81.3%), followed by the age over 60 years old(11.5%) and 0–14 years old(7.2%).

At present, the built up area of Xianyang city is mainly in Qindu District. According to the main data of the seventh national census of Qindu District (the well-developed city center of Xianyang) [1], the average population of each family is 2.42, including 49.47% males and 50.53% females. As for the age structure, the main group is 15–59 years old(66.61%), followed by the age over 60 old(17.89%) and 0–14 years old(15.5%).

Comparing the above two groups of data, it can be obtained that the survey sample is basically consistent with the Seventh Census Data in average household size and gender ratio. Due to the impact of safety and physical strength on cyclists, people aged between 18 and 45 are the main force of cycling[2]. The age distribution of the sample is consistent with that of the actual bicycle users. Therefore, our sample is consistent with the actual situation of urban citizens.

[1]The main data of the seventh national census in Qindu District http://www.snqindu.gov.cn/html/zwgk/xxgkml/zdlyxxghk/tjxx/202106/47607.html

[2] iimedia Research. 2018 Special research on the development status of Shared bikes in China. https://www.iimedia.cn/c400/63243.html.2018

2. Figure 2-figure 10. These nine figures are just a simple list in the main body of the paper. It is suggested to add corresponding discussion and explanation.

Response:

Thanks for your valuable suggestion.

We have added the discussion and explanation of Figure 2-figure 10 in the revised manuscript according to your suggestion.

As can be seen from S2, the population density gradually decreases from the center to the periphery, and the population density in the areas with dense commercial service facilities is higher than that in the new urban areas. It can be seen from S3 and S4 that both the POI diversity and POI density show a typical pattern of contiguous distribution and spread along the line. Contiguous distribution mainly includes the old street area, the business circles on Maotiao Road, the business circles of 505 Square an others, while the spread line mainly includes Renmin Road, Yuquan Road, Weiyang Road, etc. As for the new urban areas, both the diversity and density of POI are low.

Seen from S5, S6, S7 and S8, the densities of road network, intersection, bus station and public service facility around the old urban areas and commercial core areas are higher than those of the urban fringe areas and newly built areas. S8 and S9 show that the traffic volume of the commercial core areas is large and most of the roads are equipped with motor and non-motor vehicle separate railings. The roads of the new areas in the west and south were built under higher construction standards with cross section of three or four boards roads in the main and secondary roads, therefore, the motor and non-motor vehicle separate railings account for a relatively high proportion. From S10, it can be seen that the proportion of non-motor lane parking in the western part of the built-up area is more serious than that in the eastern part. For one thing, there are relatively few traffic monitoring facilities in the west and the law enforcement is weaker than that in the east. For another, there are fewer public parking lots in the west areas compared with the east areas.

The modified contents have been marked in red font on Page 11 line 237-241, Page 12 line 242-256 in the updated manuscript.

3. This paper constructs three models, but the construction premise of the three models is not clear, and the relationship and research value of the three models are not explained in detail.

Response:

Many thanks for pointing out this issue.

We have supplemented and improved the construction premises, interrelation and research values of these three models in detail.

1、The construction premises of the three models

Traditional linear models, such as variance analysis model and regression analysis model, can only analyze problems involving one layer of data, but cannot comprehensively analyze problems involving nested and hierarchical data. In many studies, the sampling often comes from different levels and units, which brings many cross-level (multi-layer) research problems, thus the multi-layer linear models can be used to effectively and statistically to solve these problems.

A random effect variable can be added to the multi-layer model to make the multi-layer model work well for internal correlation data and data with unequal variance. The multi-layer model assumes that the level 1 residual follows a normal distribution and the level 2 residual follows a multivariate normal distribution, and that the level 1 and level 2 residuals are independent of each other.

2、the relationships of the three models

At the beginning of the study, no one knows for sure which model could fit the data most satisfactorily, could provide meaningful and interpretable results and could be a parsimonious model at the same time. Generally speaking, modeling is usually an exploratory process based on both statistical and theoretical considerations. Similarly, modeling a multi-layer model is also a step-by-step process. Combining the methods recommended by Hox (1994) and Singer (1998)[3,4], this research firstly fit an empty model to examine the outcome variability. Then, it sequentially examined the macro and micro explanatory variables that may affect the outcome variability and their cross-level interactions.

The empty model is the basis of modeling a multi-level model, because it can provide an estimate of the correlation coefficient within the data group so as to judge whether it is necessary to model a multi-level model. It is only necessary to continue the modeling of a multi-layer model after determining that there is a significant correlation within the data group. Therefore, it is always the first step to run the empty model in the process of modeling a multi-layer model.

The next step in modeling a multi-layer model is to add level-1 (Hox, 1994) or level-2(singer, 1998) explanatory variables into the empty model. We prefer the approach of Singer (1998) to extend the model by adding level-2 explanatory variables. If the results of the empty model show that the data has significant within-group correlation or within-group homogeneity, it means that the data has between-group heterogeneity. Thus, the between-group variation in the mean outcome remains to be explained. Logically speaking, the next step in modeling is to include group-level variables in the model to explain this variability.

The individual characteristic, that is, the level-1 explanatory variable was not controlled when examining the relationship between the group outcome means and the group-level variables. In the third step of modeling, we added level-1 variables to the model and treated all the slopes of level-1 variables as fixed. The set models in the modeling steps 2 and 3 have the same random effects, but different fixed effects.

After the third step of the modeling exploration, we determined which slopes of the level-1 variables in the model are random coefficients in order to test the complete model with both level-1 and level-2 variables. The complete model can account for or explain how the overall outcome variables are affected by the first and second layer factors with theoretical construction. Also, the slopes of level-1 variables can used as a function of the group level explanatory variables in the macro model for cross-layer interaction to explain the between-group variation at a group level.

2、Research value

First of all, the traditional linear model can only analyze the problems involving a certain level of data, rather than comprehensively analyze the problems involving two or more levels. The multi-level linear model is an effective statistical method to solve these problems. This model can not only study the effect of individual level and group level variables on outcome measurement, but also provide an opportunity to study the variation of outcome measurement at different levels.

Secondly, the multi-layer linear model can deal with missing values on the basis of maximum likelihood or restricted maximum likelihood estimation. Therefore, the original data do not need to remove the research objects with missing values, nor need to make up for the missing observation values.

Finally, the multi-layer linear model does not require the observation values to be independent of each other, so it can correct the parameter standard error estimation bias caused by the non-independent observation data. The parameter estimation method of the model is more stable and the prediction results are more effective.

[3]Hox, J. J. 1994. Applied Multilevel Analysis. Amsterdam. TT-Publikaties.

[4]Singer, J. D. 1998. Using SAS Pro Mixed to fit multilevel models, hierarchical models, and individual growth models. J. of Educ. Behav. Stat. Vol.24: 323-255.

The modified contents have been marked in red font on Page 13 line 279-288 in the updated manuscript.

4. The innovation and practical application scenarios of the paper need to be further elaborated.

Response:

Many thanks for pointing out this issue.

The innovation of this research was added at the beginning of the sixth paragraph and How the study is useful for practical purpose was added at the end of Conclusion as one paragraph.

At research perspectives aspect, this research focuses on the travel generation prediction of cyclists instead of all-mode resident travel generation prediction model. It is committed to studying the correlation mechanism between the per capita cycling frequency of household, household attributes and built environment elements, which promotes green and healthy lifestyle.

As far as research elements are concerned, the built environment elements of existing research mainly include population density, traffic facility density, POI diversity and other conventional indicators. This research has not only included two variables of motor and non-motor vehicle separate railing ratio and non-motor lane parking ratio that affect cycling safety, but also included some factors that are closely related to daily life and have a significant impact on the per capita cycling frequency of household, such as the density of education, shopping, leisure and medical facilities.

From the aspect of research methods, this research has broken through the defect of losing data information in the disaggregating or aggregating process of traditional regression models. According to the nested structure of family-community data and considering the within-group homogeneity and between-group heterogeneity of the data, a multi-level linear regression model was used in this research to explore the effects of household and community level variables on the per capita cycling frequency of household and the cross-layer interaction effects between household and community.

To sum up, diversity of public service facilities, good road infrastructure and appropriate policies are conductive to the cycling safety and comfort for community members to promote cycling. Same specific measures are as follows: firstly, on the urban construction, reasonable arrangement of education, medical and other facilities adjacent to home areas can ensure that daily travel distances of residents are within a proper range of cycling. Secondly, it is necessary to improve the safety of cycling environment by establishing bicycle lanes and other facilities. Finally, if the existing built environment is difficult to optimize, policies such as increasing the barrier to purchasing cars shall be considered, so as to curb the use of motor vehicles and promote active travel behavior.

The modified contents have been marked in red font on Page 5 line 110-112, Page 6 line 113-119 and Page 27 line 552-560 in the updated manuscript.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

19 Apr 2022

Built Environment Impact on the Per Capita Cycling Frequency of Family

PONE-D-21-40172R1

Dear Dr. Zhang,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Quan Yuan, Ph.D.

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: My comments have been welll addressed, and thus I recommend the editor to accept the wondoerful manuscript.

Reviewer #2: The author has carefully revised the paper according to the review opinions. I recognize the research work and innovation of the paper, and suggest that it can be published in this journal.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

4 May 2022

PONE-D-21-40172R1

Built Environment Impact on the Per Capita Cycling Frequency of Family ——Based on Two-level Hierarchical Linear Model

Dear Dr. Zhang:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Quan Yuan

Academic Editor

PLOS ONE