Data sources
Individual-level data: the STEPS survey 2015 was a cross-sectional survey applying the methods and tools of the WHO STEPS. The WHO Stepwise approach to surveillance (STEPS) is a simple, standardized method for collecting, analyzing and disseminating data in the World Health Organization (WHO) member countries[12]. The survey was conducted in all 63 provinces and cities of Vietnam, between June and October 2015.
Provincial-level data: data on provincial characteristics were extracted from two surveys: (1) The 2014 Intercensal Population and Housing Survey (a large sample survey implemented in the middle of two Population and Housing Censuses) and (2) The VLSS 2014.
Individual and provincial data files were linked together using the unique provincial identities (ID) in the 2009 Vietnam Population and Housing Census.
Sample size
The sample size was calculated using the WHO’s sample calculator for STEPS. The samples were stratified by gender and three age groups (18–29, 30–49, and 50–69). The minimum sample size for each stratum was calculated according to the following formula:
In which z = 1.96, e = 0.05, p = 0.5. Then considering design effect of 1.5, an expected response rate (80%), and the number of stratum age/sex groups (6), the total sample estimated was 4,320. With the response rate of 79.5% for all three STEP rounds, the final sample consisted of 3,068 people aged 18–69 years old residing in Vietnam at the survey time.
Sampling method
A 2-stage random systematic sampling method was used. The sampling frame for the survey was developed by the General Statistics Office of Vietnam (GSO) based on the master sampling frame used in the Population and Housing Census 2009 and updated with the 2014 data [13]. Based on the Population and Housing Census data 2009, the GSO used 15% of the master sample as the future national sampling frame for the STEPs survey. The master sample contains 25,500 enumeration areas (EAs) from 706/708 districts of Vietnam (2 island districts were excluded from the GSO master sample frame). Besides, we employed a selection probability proportional to size (PPS) sampling method, where the size calculated by dividing the selection probability of an EA from the entire target population by the selection probability of an EA from the master sample. The master sample frame of GSO was stratified by two variables: urbanization (1 = urban; 2 = rural) and district group (1 = district/town/city of province; 2 = plain and coastal district; 3 = mountainous, island district). In other words, it consisted of 6 sample frames (or strata). In the first stage of the sampling process, 158 EAs in the urban areas were selected for the survey, equating the number of those in the rural areas.
The second stage of sampling involved the selection of 10% of households in each EA. Thus, 15 and 14 households were respectively selected from these urban and rural EA, using a simple systematic random sampling method. Finally, STEPS 2015 recruited 4,651 households, each of which had one eligible member automatically selected for the STEP 1 interview by the PDA program. The probability of selection was the product of the probabilities of selection for each stage. The sampling base weight of an eligible individual was the inverse of this selection probability.
Measurement
Study outcomes
Total MET score: In the STEP survey in 2015, PA-related data were collected using the Global PA Questionnaire Version 2 (GPAQ–2)[14]. The GPAQ–2 contains 16 questions on frequency (days) and duration (minutes/hours) of moderate and vigorous-intensity PA in three domains for a typical week, namely work, transport, and recreation. The data were then converted into energy expenditure measured by Metabolic Equivalent Tasks (METS, a ratio between the working metabolic rate and the resting metabolic rate) following the instruction in the GPAQ–2[14]. The total MET score is the sum of all METS/minutes/week from moderate-to vigorous-intensity PA in three domains.
Not meeting the WHO’s PA recommendations: According to the WHO, “throughout a week, including activity for work, during transport and leisure time, adults should do at least an equivalent combination of moderate- and vigorous-intensity PA achieving at least 600 MET-minutes”. Therefore, a person with a total MET score of <600 MET-minutes in this study was defined as “not meeting the WHO’s PA recommendations”
Independent variables
Subject-specific characteristics (Level 1)
The following demographic variables of the survey participants were evaluated: 1) gender (i.e., male/female); 2) age group (i.e., 18–29; 30–49; and 50–69 years old); 3) educational level (i.e., less than primary education, primary school, middle school, high school and university/college and higher); 4) current primary occupation (i.e., farmer, government staff and others, including housewives, small traders, temporary workers, housekeepers, handicraft makers and jobless people, etc.); 5) type of residential area (i.e., urban and rural); and 6) household economic status. We measured household economic status based on an asset-based wealth index, constructed them using principal component analysis (PCA), and divided them into quintiles (i.e., lowest, lower middle, middle, upper middle, and highest). Previous studies had suggested the use of PCA to create a proxy indicator of SES from the above items in the Vietnamese context [15].
Province-specific characteristics (Level 2)
The socio-economic characteristics of 63 provinces were measured using two variables: (1) Provincial monthly income per capita: the average per capita income from all sources in 2014; and (2) Provincial urbanization rate: the percentage of the population living in urban areas in the province in 2014. That information was extracted from the VLSS 2014 [16].
Statistical approach
This study employed the approach to multilevel data analysis developed by Bryk and Raudenbush’s [17]. This approach enabled us to collect data on the direct effects of individual and provincial-level factors on the total MET score and adjusted standard errors due to the effects of clustering of subject-specific measures within provinces. We designed three models as follows: 1) the first model or empty model had with no explanatory variables; 2) the second model included all individual-level variables; and 3) the third model examined variables at both individual and provincial levels. The data were extracted from the census with survey weight. We used MLwiN to analyze both scaled weighted data and unweighted data, however, weighted and unweighted data did not diverge significantly in general. To examine the contribution of provincial factors to individual total PA score, a null linear model (without any individual or school factors) was fitted, in which the provincial variance was statistically significant and accounted for 19.5% of the total variability in individual total PA scores. Adding all individual variables to the model helped to reduce the variance at the provincial level to 13.9%. In the model incorporating all individual and provincial factors (presented in table 3), the variance at the school level still stayed significant but now only accounted for 11.8% of the variability in the outcome.