Income and Wealth in the National Longitudinal Surveys By Jay Zagorsky July 2007.

Presentation on theme: "Income and Wealth in the National Longitudinal Surveys By Jay Zagorsky July 2007."— Presentation transcript:

1 Income and Wealth in the National Longitudinal Surveys By Jay Zagorsky July 2007

2 Plan For This Talk Why do you care? Basics ideas in wealth and income. What wealth and income data exist in NLS? What summary variables exist? Topcoding Problems to watch out for Conclusions

3 Why Income and Wealth? Important predictors of a host of socioeconomic outcomes (e.g., education, occupation, family behavior, delinquency, etc etc etc) Income and wealth are also important outcomes themselves. They are the basis of most inequality, and we still do not completely understand why some people have more income and wealth than others.

4 What You Should Find When using income and wealth variables past research has found: –Asians tend to have highest incomes, white non- Hispanics 2 nd highest, Hispanics 3 rd and blacks 4 th –Whites have the highest wealth, Asians 2 nd, Blacks and Hispanics have very low wealth –Both income and wealth are predicted by usual suspects (education, family background, etc) and some less-studied predictors (religion, high school work)

5 Basic Wealth Ideas 2 Types of Wealth data –Assets: Items which have value and can be resold to others. Examples, cars, homes, stocks and bonds. –Debts: Money owed by respondent to pay for assets or for past consumption. Examples, Credit Card balances, mortgages, and car loans. Most research focuses on net worth Net Worth = Assets – Debts Why net worth? Some people who live in million dollar homes and drive Mercedes have net worth close to zero. How rich or poor a person is not a function just of assets or just debts but both.

6 Basic Income Ideas 2 Types of Income data –Earned: Money which is tied to effort or hours on the job. Examples, wages, salary, tips. –Unearned: Money which is not tied to amount of work. Examples, Stipends, scholarships, unemployment insurance, dividends, and interest. Most research focuses on total income Total Income = Earned + Unearned Why total income? Income is key measure of how much an individual can buy/consume. Some people have very low earned income (retirees) but high unearned. Others have low unearned but high earned (new job market entrants).

7 Biggest Basic Idea Income and wealth while correlated do not track same underlying socio-economic concepts. –Correlation of NLSY79 total income and net worth data in 2000 survey is just 0.41. Some people with high incomes have almost no wealth –I met a surgeon who doesn’t bother to save. “Why should I? I can always earn as much as I want.” Some people with low incomes have large wealth. –Retirees in Florida and Arizona who are spending down their savings have low income but live quite well. Some people have both. –Low income and low wealth –High income and high wealth

8 NLS Income and Assets Details Next slides answer –When were questions fielded? –What type of information is available? –What summary information is available? –What should a user look out for?

9 Assets

10 Assets Questions Were Fielded NLSY79 Part of many but not all NLSY79 questionnaires: –Questions were asked in 1985, 1986, 1987, 1988, 1989, 1990, 1992, 1993, 1994, 1996, 1998, 2000 & 2004. –Questions not asked prior to 1985 –Not asked in 1991, 2002 & 2006 surveys

11 Assets Questions Were Fielded NLSY97 Questions appear in every survey but not asked to every respondents. –Only asked when a respondent becomes “independent” (married, or had child, or turned 18) –When respondent turns age 20 –When respondent turns age 25

12 Assets Questions Were Fielded Original Cohorts & NLSY79Child Mature Women –Questions asked just in 1967, 1971, 1972, 1977, 1982, 1987, 1989, 1995, 1997, 2001, and 2003. Young Women –Questions asked just in 1968, 1971, 1972, 1973, 1978, 1983, 1988, 1993, 1995, 1997, 2001, and 2003. Young Adults –1994, 1996, and 1998.

13 High Level Asset Information Two types of high level information: ownership and amount. Ownership example: “Is this (house/apartment) owned or being bought by [your/you/you or your] [spouse/partner]?” –Results in Yes / No Amount example: “About how much do you think this property would sell for on today's market?” –Results in $ amounts

14 Low Level Asset Information Examples of ownership –Own Home/Apartment –Have Estate/Trust –Have Savings Bonds –Have Stocks/Bonds –Have IRA/Keogh/401k –People Owe You Money –Own Farm/Business/R.E. –Own Vehicles –Owe Money to Creditors Examples of amount –Value of Property –Value of Estate/Trust –Value Savings Bonds –Value Stocks/Bonds –Value IRA/Keogh/401k –Amount People Owe You –Value Farm/Business/R.E. –Vehicle Value –Debts What Specific Questions Are Asked When?

15 See NLS User Guides For Details

16 Refusals and Don’t Knows In early years respondents who refuse or don’t know are marked as –1 or –2. Newer surveys (chiefly 79) use two approaches –Range. Ask respondent for an upper and lower bound. Many respondents give quite precise bounds. Simplest way to handle range is to take midpoint of upper and lower. –Unfolding brackets. Respondents are asked if amount is above or below a cutoff (I.e. vehicles worth more than $5,000). As long as their answer is not a refusal or DK they are then asked one more above or below question. Either vehicles worth $1,000 or more OR vehicles worth $10,000 or more? –This results in 3 defined ranges and 1 open ended range.

17 Summary Asset Information Mature Women has Total Net Family Assets created for 1967, 1971, 1972, 1977, 1982, 1987, 1989, and 1995- 2003 surveys. Young Women has Total Net Family Assets, created in 1968, 1971–73, 1978, 1983, 1988, and 1993–2003 surveys. NLSY97 has created variable net worth variable called CV_HH_NET_WORTH_Y for all surveys from 1997 to 2004. Also has a series in 1997 of net worth based on parent’s answers. Currently no created net worth variables for Young Adult cohorts. NLSY79 will soon have created net worth. Caution: some created net worth values are negative

18 NLSY79 Wealth Summary When the NLSY79 summary variable is released it will include imputed variables in the summary for items where R stated don’t know or refuse. Two imputation methods were used. –Linear interpolation. If respondent states their car was worth $5,000 in 1988, don’t know in 1989 and $7,000 in 1990, then the 1989 value is estimated as $6,0000. –Linear regression. Interpolation does not work for missing end points (i.e. latest survey R says don’t know) so OLS regression is done to estimate missing end point numbers.

19 Cohort Differences NLSY97 has deep focus on housing. –Separate questions for mobile homes, farms –How much did you make from selling last residence –Date housing purchased –Amount $ spent improving housing –Cost of utilities –Who loaned money to R to purchase housing

20 Cohort Differences NLSY79 has questions on R’s perceptions of wealth –Allow you to compare calculated Net Worth with perceived net worth. –Perceived net worth is an excellent instrument variable. YA questions almost carbon copy of NLSY79 but YA Grant only asked a stripped down set of questions. YW in early years have fewer questions than MW.

21 Income

22 Income Questions Were Fielded Key idea – Some kind of income information is available in every survey for every respondent. NLSY97: All years 1997 to 2003. NLSY79: All years 1979 to 2004. Young Adults: All years 1994 to 2004 –Note non-emancipated youths (respondent living at home, or who are < 18) are asked fewer questions for above 3 cohorts. Mature Women Young Women –Income questions asked in all face-to-face surveys. Income questions greatly pared down in mail surveys. (ex. MW 1968)

23 How Are Income Sections Setup? Typical NLS income section contains four types of questions. 1.First part determine income from wages, salaries and tips, and self-employment. 2.Second part determines government transfers and welfare payments. 3.Third part deals with private transfers such as child support, alimony and gifts. 4.Last respondents lists income from other sources such as scholarships, interest, dividends and rent.

24 High Level Income Information Two types of questions: Separate vs. Combined. Separate Questions: Respondent is asked for their information and then information about spouse/partner. (examples from YA surveys) –“During 1999, how much did you receive from wages, salary, commissions, or tips from all jobs before deductions for taxes or anything else?” –“During 1999, how much did your [spouse or partner] receive from wages, salary, commissions, or tips from all jobs before deductions for taxes or anything else?” Combined: One question covers everyone –“What was the total amount of these veterans benefits, worker's compensation, disability payments, or payments from Social Security [your/you/you or your] [spouse/partner] received during 1995?”

25 Low Level Income Information Examples of questions asked –Wage and salary –Military pay –Business income –Education Benefits –Food Stamps –Unemployment Compensation –AFDC / TANIF / Welfare Payments –Child Support –SSI –Disability / Workers Compensation –Interest / Dividends –Rental Income What Specific Questions Are Asked When?

26 See NLS User Guides For Details

27 Summary NLS Income In each round is a created variable which combines all the income. Often called “Total Net Family Income” Calculated by summing everything. Does not include –Partner income –No items are subtracted from the 'Net Family Income' variables to account for taxes or other adjustments Does include –Food stamps (Census Bureau in their income calculations does not include food stamps). If any variable is missing entire series set to missing –Results in many Don’t Knows. (2002 NLSY79 - 1,015 DKs)

28 Interesting Income Data NLSY79 has month-by-month series that track amounts received from various government programs. –Researcher can see exact $ history of AFDC or Food Stamp payments. Original cohorts asks respondent to subjectively evaluate their income. NLSY79 and NLSY97 have created variables which show if family is under official poverty line. –This variable is based on family size and year (line changes each year). Difficult to create on your own. Done for you.

29 Topcoding

30 The highest wealth and income data values are blurred and replaced with different information. –This blurring is called topcoding. Topcoding is designed to prevent respondent identification. Two methods of two coding used. –Hard cutoff: All people above a particular value are topcoded. –Top Percentage: All values in right hand tail are topcoded.

31 Topcoding II Hard cutoff: –Primarily used in original cohorts and early years of NLSY79. –Example: Mature Women’s home and farm values. All people who reported figures above a million dollars were given $999,997. –NLSY79 All income values from 1979 to 1984 above $75,000 were set to $75,000. Top Percentage: –Original cohorts combined top 3 values and replaced these 3 with the averages. –NLSY97 and NLSY79 combined the top 2% of values and replaced values with average of top 2%.

32 Topcoding III What does this mean for your work? –Calculated mean of wealth or income is lower than “true” average when hard cutoff is used. –Calculated mean of wealth or income is identical to “true” average when top percent cutoff is used. –You cannot use NLS data to understand people who have very high income or very high levels of wealth. –You can use NLS data to understand people with high, middle and low income and wealth levels.

33 Issues to Know About - Assets Nonresponse –Very high response rates to ownership questions –High but not very high response rates to amount questions Small accounts are missed since many of the questions are quite broad. Many respondents truly do not know an accurate value for many accounts (ex. amount in 401K or 403B plan). Assets with uncertain values are not asked about. –Key missing question is stock options. Wealth is highly concentrated in U.S.A. NLS over samples poor, not rich, so survey gives exceptional detail of left hand side of the income and wealth distribution NOT the right hand side.

34 Issues to Know About - Income Total wage and salary figure does not match earnings computations from event histories for many respondents Wage and salaries of spouse/partner are often not estimated correctly Respondents are sometimes confused about whether they should report pre-tax or post-tax income.

35 New Items NLSY79 2004 Asset section was redesigned –New questions on bankruptcy, financial difficulty –New questions track specific types of retirement accounts and how much Respondent or Spouse borrowed from these retirement accounts. –New questions track amount R owes for student loans, both for themselves and for their children NLSY79 2006 has consumption questions. NLSY79 will be releasing shortly –Net Worth series as part of the public release –Cleanup of amount values from 1985 to 2004 that are currently marked as out-of-range but should have been marked as don’t know.

36 2 Take Aways NLS income and wealth data have relatively low refusal & don’t know rates Survey benchmarks well against other large national surveys (Decennial Census and CPS for income, and SCF for wealth). Place NLS does not match well is in upper right hand tail (very rich).

37 Research Ideas Many basic relations are still untapped: high school behavior/young adult behavior  adult outcomes Family background effects on adult wealth and income have not been fully exploited Wonderful, longitudinal data can be used to understand process and patterns over time Sample continues to age: continues to provide rich source of information on a changing sample

38 Conclusions Income and Wealth data are two key socio- economic series that add another dimension to your research. They are likely to be important regardless of your question There are very useful resources available for understanding the data – consult it!