Supporting Online Material for

Transcription

1 Supporting Online Material for Objective Confirmation of Subjective Measures of Human Well-being: Evidence from the USA Andrew J. Oswald* and Stephen Wu *To whom correspondence should be addressed. This file includes: Information about Data and Methods Tables S1 to S3 Figures S1 and S2 References

2 Supporting Online Material for Science Manuscript Andrew J. Oswald,1* Stephen Wu,2 1Departments of Economics, Warwick University, Warwick, UK 2Departments of Economics, Hamilton College, Clinton, NY 13323, USA To whom correspondence should be addressed; Data The Behavioral Risk Factor Surveillance System (BRFSS) was established in 1984 by the Centers for Disease Control and Prevention (CDC); currently data are collected monthly in all 50 states, the District of Columbia, Puerto Rico, the U.S. Virgin Islands, and Guam. The data are designed to identify emerging health problems, establish and track health objectives, and develop and evaluate public health policies and programs (1). More than 350,000 adults are interviewed each year, making the BRFSS the largest telephone health survey in the world. All data, documentation, and codebook information can be downloaded at the CDC s website: The survey contains standard demographic information including age, race, gender, education, and marital and work status. In addition, the survey asks a series of health related questions, including questions about health habits and behaviors, history of medical conditions, health insurance coverage, and overall physical and mental health status. Importantly, the survey also asks a question about general life satisfaction. The exact wording is In general, how satisfied are you with your life? Here people are able to answer one of the following: Very Satisfied, Satisfied, Dissatisfied, or Very Dissatisfied [Questionnaire line code 206]. 1

3 Within the BRFSS questionnaire, individuals are asked quite early on about their days of poor mental health, Now thinking about your mental health, which includes stress, depression, and problems with emotions, for how many days during the past 30 days was your mental health not good? Then eight pages (of questions) later, they are asked to provide information about their household income (eight bracketed categories are provided, with the highest category including those with incomes greater than $75,000 per year). Twelve pages after the income question, they are asked about their feelings of satisfaction with their own life. We use four waves of the BRFSS, from 2005 to 2008, and we limit our sample to respondents between the ages of 18 and 85 with non-missing demographic and life satisfaction information. The resulting sample is slightly less than 1.5 million respondents, spread over four annual surveys. Table S1 describes the means in the data set. Roughly 60 percent of the sample is female, 8 percent and 6 percent are Black or Hispanic, respectively, and the average age of the sample is slightly under 53 years. 90 percent of the sample has at least a high school degree, while 33 percent has completed college. In the BRFSS, life satisfaction is treated in a cardinal way by assigning 1 to 4 to the four answers, where very satisfied is assigned a 4. Most individuals indicate being either very satisfied or satisfied with their lives, with a mean life satisfaction of 3.4, with a standard deviation of 0.6. Empirical Methods We use a linear ordinary least squares (OLS) estimator to predict life satisfaction with the following control variables: age, age squared, seven banded dummy variables for household income, dummy variables for gender, race, marital status, and different levels of education, and dummies for being retired, a student, unemployed, or a homemaker. Also included as independent control variables are eleven dummy variables for the month of interview, and a set 2

4 of 50 state dummies (because D.C. is included the regression, the total number of states is 51). Although our analysis uses a linear OLS estimator in which the four possible satisfaction values of the dependent variable are assigned the integers from a high of 4 down to a low of 1, the substantive findings are not altered by switching to ordered models such as probit or logit. The simpler method allows coefficient sizes to be read off directly as approximate life-satisfaction points. Table S2 describes the regression-corrected life satisfaction patterns across the geography of the United States. Here the state-dummy coefficients are written out explicitly. Alabama is the omitted, base category; it is chosen because it comes first alphabetically; the choice of base makes no difference to the analytical conclusions. The first state-dummy coefficient in column 1 of Table S2 can be interpreted as showing that corrected satisfaction with life on average in Alaska is life satisfaction points below that in the base case of Alabama; Arizona is indistinguishable from Alabama; and so on across the listed states. We use these state coefficients as the first way to assess the quality of life across different locations within the United States. For the second measure of quality of life across states, we use an index calculated by Gabriel, Mattey, and Wascher (2). They use a series of objective indicators to calculate an index for quality of life across U.S. states for the year 1990, which is the most recent estimate in the economics literature. Some of the amenity controls include weather and other climatic variables (precipitation, wind speed, sunshine), recreation opportunities (whether a state borders an ocean, the Gulf of Mexico, or the Great Lakes, the number of visitors to state and national parks relative to state population), and other state specific characteristics (violent crime rates, environmental regulation, income taxes). The full set of amenity characteristics are listed in Table S3. 3

5 These objective data were combined in a weighted average, where the weights were not chosen arbitrarily but according to coefficient sizes taken from regional hedonic price equations. In their study, they use data from the 1990 Census of Population and Housing and the American Chamber of Commerce Research Association to estimate quality adjusted wages, housing costs, and non-housing cost-of-living. These three variables are regressed as a system of equations on the set of amenity characteristics listed in Table S3 to estimate their responsiveness to changes in amenities. Then, a full implicit price is calculated for each amenity. This can be thought of a monetary amount needed to compensate a resident for something with a negative price (e.g., a disamenity such as precipitation), or the amount that one would be willing to pay for something with a positive price (e.g., an amenity such as being near inland water). Gabriel provides the example that for each additional inch of yearly precipitation, there is a compensating differential of $18. Therefore, someone living in a state with yearly precipitation of 20 inches has a quality of life index that would be $180 lower than someone living in a state with yearly precipitation of 10 inches ($360-$180=$180). To calculate quality of life indices, state averages for the various amenities are multiplied by the full implicit prices for each of the amenities. Aggregating across amenities generates a quality of life index for each state. The differences in state values can be thought of as the amount one would pay in order to live in a state with a higher quality of life index. Louisiana and Income As a general check, as noted in the main manuscript, Figure S1 shows that Louisiana also comes out strongly in a league table of mental health from an independent source. Figure S2 shows that higher-income states are not happier states. Size of coefficients 4

6 To illustrate the size of coefficients, and to give a sense of what 0.12 cardinal lifesatisfaction points means, a clip from some Stata output is attached. Sensitivity Checks Because the BRFSS data is clustered at the state level, we also have done some additional analysis using multi level models. Unfortunately, because of the size of the data and the number of independent variables in the main regression, we are unable to obtain estimates for the full model because of a lack of convergence. However, in using more parsimonious specifications, estimates using multi level models are similar to estimates obtained using ordinary least squares. In most instances, the standard errors are smaller using multi level modelling. Therefore, we are not too concerned that the clustered nature of the data is biasing our results. In addition, we have also estimated separate regressions for each year of the data to see if the state level coefficients change in any appreciable manner. While there is some variation, it is minor. The rankings of state level adjusted life satisfaction are quite stable from year to year. 5

7 Table S1: Summary Statistics for BRFSS Pooled Data: Variable Mean Std. Dev. Life Satisfaction (1-4 Scale) Income 10-15K Income 15-20K Income 20-25K Income 25-35K Income 35-50K Income 50-75K Income >75K Age Black Asian Hispanic Native American Other Minority Female Some High School High School Some College College Married Divorced Separated Widowed Partner Self employed Unemployed Homemaker Student Retired # Observations 1,483,403 6

8 Table S2: Estimates from Life Satisfaction Equations (1) (2) (3) (4) State coefficient dummy-variable values [95% confidence interval] Life-satisfaction using BRFSS data, rounded to two decimals Objective state quality-of-life rank: comp. diff. method (Gabriel et al 2003) State rank using regression-adjusted life satisfaction in BRFSS data Alaska [-.0179, -.008] Arizona [-.0024,.0053] Arkansas [-.0191, ] California [-.0802, ] Colorado [-.0305, ] Connecticut [-.0844, ] Delaware [-.0291, ] District of Columbia N/A 37 [-.0507, ] Florida [.0019,.0062] Georgia [-.0229, ] Hawaii [.0038,.0179] Idaho [-.0175, ] Illinois [-.0741, ] Indiana [-.0804, -0766] Iowa [-.0439, ] Kansas [-.0465, ] Kentucky [-.0473, ] Louisiana [.0318,.0338] 7

9 Maine [-.0089, ] Maryland [-.0689, ] Massachusetts [-.0729, ] Michigan [-.0814, ] Minnesota [-.0337, ] Mississippi [.0000,.0014] Missouri [-.0657, ] Montana [-.0025,.0040] Nebraska [-.0469, ] Nevada [-.0676, ] New Hampshire [-.0365, ] New Jersey [-.0813, ] New Mexico [-.0340, ] New York [-.0903, ] North Carolina [-.0147, ] North Dakota [-.0328, ] Ohio [-.0714, ] Oklahoma [-.0295, ] Oregon [-.0437, ] Pennsylvania [-.0689, ] Rhode Island [-.0713, ] 8

10 South Carolina [.0004,.0023] South Dakota [-.0170, ] Tennessee [.0011,.0042] Texas [-.0180, ] Utah [-.0300, ] Vermont [-.0205, ] Virginia [-.0354, ] Washington [-.0488, ] West Virginia [-.0462, ] Wisconsin [-.0399, ] Wyoming [-.0158, ] Constant Observations 1,213,992 R-squared Notes: In these state rankings, a small number indicates a high quality of life; 50 is the lowest quality-of-life rank. All regressions include controls for income, personal characteristics, month, and year of survey. Life satisfaction is measured on a 1-4 scale; 1=very dissatisfied, 4=very satisfied. Alabama is the base, omitted category; it has, by construction, a coefficient of zero. Its rank in column 3 would be 26. Its rank in column 4 would be 9. This is a normalization; any state can be chosen as base. 9

11 Table S3: Amenities Used by Gabriel et al. (2003) to Calculate Quality of Life Rankings Precipitation U.S. National Oceanic and Atmospheric Administration, Climatography of the United States, No. 81 as published in The Statistical Abstract of the United States, 1993, Table No Humidity U.S. National Oceanic and Atmospheric Administration, Comparative Climactic Data, as published in The Statistical Abstract of the United States, 1994, Table No Heating Degree Days U.S. National Oceanic and Atmospheric Administration, Climatography of the United States, No. 81 as published in The Statistical Abstract of the United States, 1993, Table No Cooling Degree Days U.S. National Oceanic and Atmospheric Administration, Climatography of the United States, No. 81 as published in The Statistical Abstract of the United States, 1993, Table No Wind Speed U.S. National Oceanic and Atmospheric Administration, Comparative Climactic Data, as published in The Statistical Abstract of the United States, 1994, Table No Sunshine U.S. National Oceanic and Atmospheric Administration, Comparative Climactic Data, as published in The Statistical Abstract of the United States, 1994, Table No Coast Specified by the authors to equal one if the state abuts an ocean, the Great Lakes, or the Gulf of Mexico. Inland Water Boating Industry magazine, January 1990 as published in Hall and Kerr, Green Index, p Federal Land U.S. Department of the Interior, Bureau of Land Management, Public Lands Statistics, 1989 as published in Hall and Kerr, Green Index, p Visitors to National Parks U.S. Department of the Interior, National Park Service, National Park Service Statistical Abstract, 1989 as published in Hall and Kerr, Green Index, p Visitors to State Parks National Association of State Park Directors, Annual Information Exchange, April 1990 as published in Hall and Kerr, Green Index, p Number of hazardous waste sites U.S. Environmental Protection Agency, press release. as published in The Statistical Abstract of the United States, 1994, Table No Environmental Regulation Leniency Composite score on Green Policies (state policy initiatives and leadership in Congress) as tabulated by and published in Hall and Kerr, Green Index, p. 5. Commuting Time U.S. Bureau of the Census, Census of Population and Housing, 1980 and 1990 as published in the United States Summary volume of General Social and Economic Characteristics for 1980, table 238 and for 1990 as published in The Statistical Abstract of the United States, 1993, Table No

12 Violent Crime Rate U.S. Federal Bureau of Investigation, Crime in the United States, as published in The Statistical Abstract of the United States, various issues. Air Quality-Ozone U.S. Environmental Protection Agency, National Air Quality and Emissions Trends Report, various issues. Air Quality-Carbon Monoxide U.S. Environmental Protection Agency, National Air Quality and Emissions Trends Report, various issues. Student-teacher ratio The numerator is public elementary and secondary school enrollment from U.S. National Center for Education Statistics, Statistics of State School Systems and Digest of Education Statistics, as published in The Statistical Abstract of the United States, various issues. The denominator is number of public elementary and secondary school teachers from National Education Association, Estimates of School Statistics, as published in The Statistical Abstract of the United States, various issues. State and local taxes on property, income and sales and other U.S. Bureau of the Census, Governmental Finances, various issues. State and local expenditures on higher education, public welfare, highways, and corrections U.S. Bureau of the Census, Governmental Finances, various issues. Cost-of-living American Chamber of Commerce Researchers Association, ACCRA COST OF LIVING INDEX serial, issues for third quarters of 1981 through

13 Some STATA output to illustrate coefficient sizes on unemployment, etc Source SS df MS Number of obs = F( 93, ) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = lsatisfy Coef. Std. Err. t P> t [95% Conf. Interval] inc inc inc inc inc inc inc age agesq e black asian hisp nativeam other female somehs hs somecoll coll married divorced separated widowed partner self_emp unemp homemaker student retired

14 Figure: S1 The Correlation Between Adjusted Mental Distress and the Proportion of Youths Aged with a Major Depressive Episode in the Past Year in NSDUH Data BRFSS Data: Sample size = 1.3 million approx. 1 y = x R= Adjusted Mental Distress % Adolescents with Major Depressive Illness NSDUH data Each dot is a state. The correlation is significant at 1% on a two-tailed test. The data on rates of adolescent depression come from Mental Health America and the SAMHSA, Office of Applied Studies, National Survey on Drug Use and Health The bottom left hand observation is Louisiana. In adjusted data, there are regression controls for the survey respondent s gender, age, age squared, education, marital status, unemployment, and race, and also year dummies and month-of-interview dummies. 13

15 Figure: S2 The Inverse Correlation Between Fully Adjusted Life Satisfaction and Median Household Income across 51 States of the USA BRFSS Data: Sample size = 1.3 million approx. y = e-6x R= Life Satisfaction Fully Adjusted (ie income also) Median Household Income (Census data) Each dot is a state. The correlation is significant at 1% on a two-tailed test. Household income data are constructed from the CPS and Census Annual Supplements. In fully adjusted data, there are regression controls for household income as well as the survey respondent s gender, age, age squared, education, marital status, unemployment, and race, and also year dummies and month-of-interview dummies. 14

16 References 1. Behavioral Risk Factor Surveillance System, Codebook Reports, Center for Disease Control. 2. S.A. Gabriel, J.P. Mattey, W.L. Wascher, Regional Science and Urban Economics 33, (2003). General References Our complementary background paper Well-being Across America: Evidence from a Random Sample of One Million US Citizens is available online through the IZA.org website (by searching on the October 2009 conference in honor of Richard Easterlin) or by contacting us directly. It describes the BRFSS data set and the micro structure of a set of detailed regression equations. These have a similar structure (U-shaped in age, monotonically increasing in income, a large negative dummy on being unemployed, etc) to that found in the now-large literature on the estimation of happiness/well-being equations. An introduction to the literature is the 1997 Economic Journal paper, called Happiness and Economic Performance, which is referenced in the Oswald-Wu Science manuscript. The more recent literature, which has become large, can be found through sources such as Andrew Clark s website at the University of Paris: search for example on Andrew Clark economics Paris. 15