«THE EFFECT OF SCHOOL FINANCE REFORMS ON THE DISTRIBUTION OF SPENDING, ACADEMIC ACHIEVEMENT, AND ADULT OUTCOMES C. Kirabo Jackson Rucker Johnson ...»
To interrogate this using the Census data, we look at all individuals born between 1955 and 1985 (the same dates of birth as our PSID sample) and between the ages of 25 and 45 (the same ages as our PSID sample). We then estimate an event-study regression on high school completion and annual personal income. That is, we regress these outcomes on a series of indicator variables denoting the year an individual turned 17 minus the year that a reform was passed in the individuals’ state of birth. As such, individuals who were 17 at the years of reforms are at 0 years of exposure, while those in the negative range were 18 or older at the time the reforms were passed in their state of birth. We present the event time plots in figure A. If there is a causal effect, we should see minimal trending for the cohorts that were too old to be exposed and improving outcomes with increased years of exposure.
Consistent with a causal effect, for both high school graduation and income, outcomes are increasing in years of exposure and there is no pre-existing trend difference for the outcomes. To provide a test of statistical significance, we take the plotted dynamic treatment effect (i.e. the treatment effect for each cohort) and then fit a simple linear model in event time with an intercept at zero (i.e. cohort that was 18 at the time of the reform) and a differential post intervention time trend (a differential trend for cohorts that were 17 or younger at the time of reforms). The F-statistic on the post reform intercept and the post reform linear trend provides a test of whether there is a statistically significant structural break that occurs for the treated cohorts. For both outcomes, the two-sided p-value associated with the F-statistic for the joint significance of the post intercept and the post trend is smaller than 0.05 – indicating that relative to the pre-existing trend in outcomes, the improved high school completion and income observed for the exposed cohorts is larger than would be expected by random change (under the simple linear model).
1500.01 1000.005 500 0 0
Data: Individual Census IPUMS Data (1970, 1980, 1990, and 2000-2012) matched with the timing of court mandated reforms by date of birth and state of birth.
Models: Results are based on event-study models that include: state of birth fixed effects, year of birth effects, age and age squared, and gender and ethnicity interacted with a linear cohort trend. The figure plots the estimated years of exposure to school finance reform for high school graduation (Right) and total personal income (Left).
72 Appendix D: Data Sources for Part 1 Per Pupil Spending Data Previous historical data on per pupil expenditures was only available in a readily usable format via the Census of Governments: School System Finance (F-33) File (U.S. Bureau of the Census, Department of Commerce). The Census of Governments previously was only conducted in years that end in a two or seven, so at the time when many important papers on SFRs were written, there were many years of missing data. In addition, until recently the earliest available Fdata was for the year 1972. As a result, it was previously impossible to model per pupil spending and spending inequality annually over time, so many authors (e.g., MES, Card and Payne), operating under the Common Trends Assumption, assumed that trends in per pupil spending were linear. Due to these limitations, previous papers on school finance reforms were also unable to look at how the exact timing of reforms affected per pupil expenditure and spending inequality within a state.
Our data from the Historical Database on Individual Government Finances (INDFIN) represents the Census Bureau’s first effort to provide a time series of historically consistent data on the finances of individual governments. This database combines data from the Census of Governments Survey of Government Finances (F-33), the National Archives, and the Individual Government Finances Survey. The School District Finance Data FY 1967-91 is available annually from 1967 through 1991. It contains over one million individual local government records, including counties, cities, townships, special districts, and independent school districts.
The INDFIN database frees the researcher from the arduous task of reconciling the many technical, classification, and other data-related changes that have occurred over the last 30 years.
For example, this database includes corrected statistical weights that have been standardized across years, which had not been done previously. Furthermore, although most governments retain the ID number they are assigned originally, there are circumstances that result in a government's ID being changed. Since a major purpose of the INDFIN database is tracking government finances over time, it is critical that a government possess the same ID for all years (unless the ID change had a major structural cause). For example, All Alaska IDs were changed in the 1982 Census of Governments. In addition, new county incorporations, where governments in the new county area are re-assigned an ID based on the new county code (e.g., La Paz County, AZ), cause ID changes. Thus, if a government ID number was changed, the ID used in the database is its current GID number, including those preceding the cause of the change, so that the ID is standardized across years.
In addition to standardizing the data, the Census Bureau has corrected a number of errors in the INDFIN database that were previously in other sources of data. For example, for fiscal years 1974, 1975, 1976 and 1978 the school district enrollment data that had previously been released were useless (either missing or in error for many records). Thus, in August 2000, these missing enrollment data were replaced with those from the employment survey individual unit files. This enables us to more accurately compute per pupil expenditures for those years. In addition, source files before fiscal 1977 were in whole dollars rather than thousands. This set a limit on the largest value any field could hold. If a figure exceeded that amount, then the field contained a special "overflow" flag (999999999). Few governments exceeded the limit (Port Authority of NY and NJ and Los Angeles County, CA are two that did). For the INDFIN database, actual data were substituted for the overflow flag. Finally, in some cases the Census revised the original data in source files for the INDFIN database. In some cases, official 73 revisions were never applied to the data files. Others resulted from the different environment and operating practices under which source files were created. Finally, some extreme outliers were identified and corrected (e.g., a keying error for a small government that ballooned its data).
The Common Core of Data (CCD) School District Finance Survey (F-33) consists of data submitted annually to the National Center for Education Statistics (NCES) by state education agencies (SEAs) in the 50 states and the District of Columbia. The purpose of the survey is to provide finance data for all local education agencies (LEAs) that provide free public elementary and secondary education in the United States. Both NCES and the Governments Division of the U.S. Census Bureau collect public school system finance data, and they collaborate in their efforts to gather these data. The Census of Governments, which was recorded every five years until 1992, records administrative data on school spending for every district in the United States.
After 1992, the Public Elementary-Secondary Education Finances data were recorded annually with data available until 2010. We combine these data sources to construct a long panel of annual per pupil spending for each school district in the United States between 1967 and 2010.
Per-pupil spending data from before 1992 is missing for Alaska, Hawaii, Maryland, North Carolina, Virginia, and Washington, D.C. Per-pupil spending data from 1968 and 1969 is missing for all states. Spending data in Florida was also missing for 1975, 1983, 1985-1987, and
1991. Spending data in Kansas was also missing for 1977 and 1986. Spending data in Mississippi was also missing for 1985 and 1988. Spending data in Wyoming was also missing for 1979 and 1984. Spending data for Montana is missing in 1976, data for Nebraska is missing in 1977, and data for Texas is missing in 1991. Where there was only a year or two of missing per pupil expenditure data, we filled in this data using linear interpolation.
Data on School Finance Reforms
Due to great interest on the topic, the timing of school finance reforms (SFRs) has been collected in various places. Data on the exact timing and type of court ordered and legislative SFRs was obtained from Public School Finance Programs of the Unites States and Canada (PSFP), National Access Network’s state by state school finance litigation map (2011), from Murray, Evans, and Schwab (1998), Hoxby (2001), Card and Payne (2002), Hightower et al (2010), and Baicker and Gordon (2004). The most accurate information on school finance laws can be derived from the PSFP, which provides basic information and references to the legislation and court cases challenging them (Hoxby, 2001). In most cases, data from these sources are consistent with each other. Where there are discrepancies we often defer to PSFP, but also consulted LexisNexis and state court and legislation records.
There were discrepancies in reported timing of overturned court cases in several states:
Connecticut (Hoxby states the decision was made in 1978, but Card and Payne report it was made in 1977), Kansas (Hoxby states 1976, but PSFP and ACCESS report 1972), New Jersey (Card and Payne state 1989, but PSFP says 1990), Washington (Murray, Evans, and Schwab, Hoxby, and Card and Payne report 1978, but PSFP reports 1977), Wyoming (Hoxby says 1983, but Card and Payne and Murray, Evans, and Schwab report 1980). We researched each case by name to discover the true date of the decision.
Using a policy survey conducted during the 2008-2009 school year, a recent study by Hightower et al (2010) provides a description of state finance policies and practices. This study was used to verify whether there had been any changes to state funding formulas between 1998 and 2009. We only collected information on the first five court cases per state in which the state 74 found the school funding system unconstitutional. There were only three states with five or more court cases overruling the funding system (New Hampshire, New Jersey, and Texas). In addition, we only collected information on the first four court cases per state in which states upheld the school funding system. There were only four states with four or more court cases in which the school funding system was upheld (Illinois, New York, Oregon, and Pennsylvania).
Information on whether or not a state funding formula had a MFP, flat grant formula, variable matching grant scheme, recapture provision, spending limit, power equalization scheme, local-effort equalization scheme, or full state funding came from PSFP (1998) and was verified using Card and Payne (2002) and Hightower et al (2010). We defined MFPs, flat grant formulas, and variable matching grant schemes in the same way as Card and Payne did in their 2002 study.
We defined power equalization, local-effort equalization, and full state funding in the same way as the EPE study (Hightower, Mitani and Swanson, 2010). Each element of a state funding formula was coded as a dichotomous variable. For example, MFP is a dichotomous variable that is equal to one in the year and all subsequent years in which a state’s finance system had a MFP plan in place. MFP was set equal to zero in all years prior to the state’s funding system having a MFP in place, or if a state never implemented a MFP. Information on adequacy and equity reforms came from Berry (2005) and Springer, Liu and Guthrie (2009)’s Table 1. Following Springer, Lui and Guthrie (2009), we define two dichotomous variables for equity and adequacy reforms. Adequacy is a dichotomous variable set to one in the year, and all subsequent years, in which a state’s finance system was overturned on adequacy grounds. Adequacy is set to zero in all years prior to a school funding mechanism being overturned, or if a state’s finance system was never ruled unconstitutional. Similarly, equity is a dichotomous variable that is set to one in the year, and in all subsequent years, in which a state’s finance system was ruled unconstitutional on equity grounds. Equity was set to zero in all years prior to the state funding mechanism being declared unconstitutional, or if a state’s funding mechanism was never overturned.
Information on the timing of tax limits came from Downes and Figlio (1998). In addition, information on the foundation property tax rate and the maximum/minimum inverted tax price of marginal local spending was obtained from Hoxby (2001) and defined in the same way. Thus, we defined a variable, limit2, which equals one if there is a zero tax price according to Hoxby, a recapture provision, or a spending cap. We defined another variable, g1taxprice, as equal to one if the inverted tax price was greater than one (which should promote spending).
Notes: The state funding formulas may include: flat grants (FG), minimum foundation plans (MFP), equalizations plans (EP), local effort equalizations (LE), spending limits (SL), and full state funding (FS).
83 Appendix E: Effect of a Reform Induced 20 Percent Spending Increase With 90 Percent Confidence Intervals
.2.1.1 0 0
1.5.5 1 0.5 0
Data: PSID geocode Data (1968-2011), matched with childhood school and neighborhood characteristics. Analysis sample includes all PSID individuals born 1955-1985, followed into adulthood through 2011. (N=15,353 individuals (9,035 poor kids; 6,318 non-poor kids) from 1,409 school districts (1,031 child counties, 50 states).
Models: Results are based on non-parametric event-study models that include: school district fixed effects, racespecific year of birth fixed effects, race*census division-specific linear cohort trends, controls at the county-level for the timing of school desegregation*race, hospital desegregation*race, roll-out of "War on Poverty" & related safety-net programs (community health centers, county expenditures on Head Start (at age 4), food stamps, medicaid, AFDC, UI, Title-I (average during childhood yrs), timing of state-funded Kindergarten), controls for 1960 county characteristics (poverty rate, percent black, education, percent urban, population size, percent voted for Strom Thurmond in 1948 Presidential election*race (proxy for segregationist preferences)) each interacted with linear cohort trends, and controls for childhood family characteristics (parental income/education/occupation, mother's marital status at birth, birth weight, gender). Standard errors are clustered at the childhood county level.
Main school finance reform variables allowed to affect outcomes through both the amount of induced school spending changes and the duration of school-age years of exposure to reform-induced spending changes (i.e., models include intercept and slope terms of intensity of treatment (district spending change) and interaction terms of "school spending change*exposure years" in order to capture dose of treatment in terms of both an individual's school-age years of exposure to school finance reform and the district's change in per-pupil spending induced by reform). Results for non-poor kids not statistically significantly different from zero.