Table 3.

Results - observational studies identified in Stage 2 that used alternative study designs or methodological approaches to support causal inference (n = 6)

Study details		Description of variables					Results (for two different methods of analysis, when reported)
		Independent variables			Dependent variables		Main method of analysis:			Alternative method of analysis:
							Panel data, RCT or SEM			Cross-sectional analysis
First author, date, journal	Study population	Description	Time varying	Areal unit	Description	Source	Description of study design	Data type (time periods)	Effect sizes (95% confidence interval) 1	Method	Effect sizes (95% confidence interval) 1
											Results where no statistically significant differences are observed between main and alternative analyses	Results where a mismatch between results is observed 2
Franzini, 2009, Am J Public Health [39]	U.S. children (all States; 10–12 year olds)	Traffic levels, physical disorder, residential density and land use	N/A	Individual Systemic Social Observations	BMI	Interviews with students and their parents, 2003	Structural equation modelling (SEM)	Cross sectional (1)	0.03 (−0.40, 0.46) (these results relate to physical activity z-scores which contributed to the SEM. Physical environment had no significant impact on physical activity or BMI in the model)	Not reported
Gibson, 2011 [40], Am J Public Health	U.S. young people (all States)	Five measures relating to food environment, including:	No	Zip-code level	BMI (obesity likelihood was also reported)	NLSY, 1998-2004	Fixed effects panel data analysis	Longitudinal data (2)	Change in BMI:	OLS	None	Under-estimates:
		(a.) supermarkets per square mile							(a.) -1.98* (−1.94,-2.02)			(a.) -0.04 (−0.18, 0.10)
		(b.) small grocery stores, and per square mile							(b.) -0.15* (−0.33,0.04)			(b.) 0.02 (−0.00, 0.04)
		(c.) full-service restaurants per square mile							(c.) 0.20* (0.03, 0.36)			(c.) -0.00 (−0.01, 0.01)
Kapinos, 2011 [38], Journal of Adolescent Health	U.S. undergraduate students (a single university campus)	Characteristics of dormitory accommodation:	No	Specific to the location of the dormitory accommodation	Weight (kg) (other outcome relating to exercise frequency, meals and snacks are not reported here)	Individual-level survey instrument (39 questions)	Randomised experiment (undergraduates were randomised to different dormitory accommodation)	Cohort data (2) One-year follow-up	Male (M) and female (F) participants:	Not reported
		(a.) on-site dining hall							(a.) M: 0.19 (−2.37, 2.76) F: 0.85* (0.12, 1.57)
		(b.) distance to gym							(b.) M: -0.25 (−1.37, 0.87) F: 0.13 (−0.32, 0.59)
		(c.) distance to central campus							(c.) M: -0.08 (−0.80, 0.63) F: -0.45 (−1.15, 0.25)
Kling, 2004, National Bureau of Economic Research [37]	U.S. (five cities; families with children; 85% with African-American or Hispanic female as household head)	Moving from a high poverty (public housing area) to a low poverty (a census tract with a poverty rate of less than ten percent) neighbourhood	No	Poverty rate was measured at the census tract level	Obesity likelihood	Individual-level survey	Randomised experiment: (moving to low poverty areas)	Cohort data (2) Five-year follow-up	(a.) intent-to-treat effect i.e. effect of being offered a housing voucher or the average effect of an attempted policy intervention on the entire target population:	Not reported
									−0.048* (−0.091, −0.005)
									(b.) treatment-on-treated i.e. those who moved using voucher
									−0.103* (−0.195, −0.011)
Powell, 2009, Journal of Health Economics [41]	U.S. young people (all States)	Measures included:	No	County level	BMI	NLSY, 1997-2000	Fixed effects panel data analysis	Panel data (4)	No statistically significant results observed in any of the measures. e.g.:	OLS	No statistically significant results observed except in one case (see right). e.g.:	Over-estimate in one case:
		(a.) restaurants per 10,000 people,							(a.) -0.03 (−0.09, 0.02)		(a.) 0.03 (−0.03, 0.09)
		(b.) grocery stores per 10,000 people							(b.) -0.03 (−0.11, 0.05)		(b.) -0.0074 (−0.10, 0.08)
		(c.) physical activity facilities per 10,000 people							(c.) -0.12 (−0.2, 0.05)		(c.) -0.16* (−0.30,-0.02)
Sandy, 2009, National Bureau of Economic Research [42]	U.S. young children (Indianapolis, Indiana)	Twenty different measures,3 including:	Yes	Individual addresses	BMI (z scores)	Clinical records, 1996-2006	Fixed effects panel data analysis	Panel data (10)	In general, very few statistically significant results3	Cross-sectional OLS	In general, very few statistically significant results.	Over-estimates in two cases3:
									However, some selected exceptions (within 0.25 miles and including children of all ages, unless otherwise stated):
		(a.) restaurants							(a.) -0.08* [−0.13 at 0.1 miles]			(a.) 0.02 [0.08* at 0.1 mile]
		(b.) supermarkets							(b.) 0.05 (0.1 miles)			(b.) -0.19* (0.1 miles)
												Under-estimates in three cases3:
		(c.)fitness,							(c.) -2.26*			(c.) 0.25
		(d.) kickball, and							(d.) -0.08*			(d.) 0.04
		(e.) volleyball facilities							(e.) -0.90* (0.1 miles; children <8 years only)			(e.) 0.03 (0.1 miles; children <8 years only)
												All within 0.25 miles and including children of all ages, unless otherwise stated

NLSY: National Longitudinal Survey of Youth dataset.

BMI: Body mass index measured in kg/m².

OLS: Ordinary-Least-Squares.

¹ * indicates statistical significance at the p < 0.05 level.

² When compared to results in the main analysis: “Under-estimate” if statistically significant results in the main analysis were not statistically significant the cross-sectional, single equation analysis; “Over-estimate” if statistically insignificant results in the main analysis were statistically significant in the cross-sectional, single equation analysis.

³ Although 80 results were reported in total, the results reported in this table were for those variables deemed by the authors of that study to be most relevant to policy makers. Results were reported for four different sized areas/buffer zones (ranging from 0.1 to 1 mile).