The 2020 Florida Price Level Index

The 2020 Florida Price Level IndexMarch 8, 2021Jim DeweyDirector of Economic AnalysisFlorida Polytechnic UniversityThe Florida Price Level Index was established by theLegislature as the basis for the District Cost Differentialin the Florida Education Finance Program. The FPLI is acomparable wage index that represents the relative costof hiring comparable personnel among Florida’s schooldistricts. The calculation is based on wage data forhundreds of occupations across Florida’s 67 countiescollected by the Florida Department of dCalhounCharlotteCitrusClayCollierColumbiaDadeDe HillsboroughHolmesIndian y’s Bureau of Labor Market Statistics as partof the U.S. Bureau of Labor Statistics’ OccupationalEmployment Statistics survey. The table below presentsthe 2020 FPLI, along with the 2019 and 2018 indices. aloosaOkeechobeeOrangeOsceolaPalm BeachPascoPinellasPolkPutnamSaint JohnsSaint LucieSanta iaWakullaWaltonWashington1This report is available at http://www.fldoe.org/fefp/ andhttps://floridapoly.edu/ .0695.7394.3998.0192.81

The Distribution of the FPLIThe Florida Price Level Index (FPLI) is constructed sothat the population-weighted state average is 100,though this normalization does not impact the relativecomparison between any two counties. The medianFloridian, ranked by 2020 county FPLI, lives inHillsborough County, with an index value of 100.73. Thatis, less than half of Floridians live in counties with indexvalues greater than 100.73, less than half live in countieswith index values less than 100.73, and the rest live inHillsborough County.The map to the right displays the distribution of theFPLI across Florida. As population density increases,workers face higher housing costs, longer commutes, orboth, for which they are compensated by higher wages.Therefore, although many things affect counties’ FPLIvalues, counties that are more urban tend to have highervalues. The six counties with FPLI values of 102 or abovecontain 22.6% of the state’s population. The twenty-onecounties with index values within two percentage pointsof the state average, from 98 to 101.99, contain 55.6%of the state’s population. Twenty counties, containing17.2% of Florida’s population, have index values from 94to 97.99. Finally, 4.6% of the state’s population live inthe twenty counties with index values below tely 17% of the public-school labor bill is paidto workers without a bachelor’s degree, which are notrepresented in that sample. Moreover, the level ofincome at a given reference location is a potentiallyimportant determinant of the relative wage pattern, andpublic-school workers with a degree earn substantiallyless than the average worker with a degree.The figure on the next page presents empiricalcumulative U.S. income distributions for all publicschool workers, all non-education workers, and all noneducation workers with a bachelor’s degree. The groupof all non-education workers appears more comparableto public-school workers than does the subset with abachelor’s degree. Further analysis suggests the gain inprecision from using the larger sample available fromOES data outweighs the gain in precision fromcontrolling for other characteristics using ACS data. 3Prior to the 2003 index, the FPLI was an index of therelative expenditure required to purchase a marketbasket of goods and services, similar to the ConsumerPrice Index, albeit in a spatial context. This approach wasadopted due to the lack of suitable wage data. Thejustification for this approach was that, all else equal,wages adjust to compensate for differences in the pricesof goods and services, particularly housing.Methodological Approach 2The FPLI is a wage index comparing the cost of hiringa state average worker among Florida’s 67 counties. Itsuse in adjusting school funding assumes the relativewage pattern for school workers is well approximated bythe relative wage pattern for the state average worker.It relies on data on wages by occupation from theOccupational Employment Statistics (OES) survey, basedon a massive employer sample. Columns 1 and 2 of thetable at the end of this document present the averagenumber of occupations and employees represented byresponses to a complete OES survey by county.An alternative would be to use data from theAmerican Community Survey (ACS) that allowscontrolling for individual worker characteristics otherthan occupation, and to focus on the subset of workerswith at least a bachelor’s degree, since teachers mustpossess one. Controlling for other worker characteristicswould increase precision. However, using the ACS datawould greatly reduce the number of workers covered by23For details on the methodology see Jim Dewey (2020) Florida PriceLevel Index Methodology, available 04.For more information, see Jim Dewey, (2019) Comparing the FloridaPrice Level Index and the Comparable Wage Index for Teachers,available at .2

Statistical Smoothing Prior to adoption of the currentmethodology, in some cases otherwise similar countieshad very different FPLI values though the estimates’margins of error were large, meaning there was littleevidence that the difference was real. Statisticalsmoothing ensures similar counties have similar indexvalues unless the estimates’ margins of error provideevidence that the difference is real.To implement statistical smoothing, the relationshipbetween the initial estimate and county characteristicssuch as the size and age distribution of the populationand per capita income is used to predict index values foreach county. This predicted value and the initialestimate are combined by taking a weighted averageaccording to their precision. The weights are calculatedto minimize the margin of error of the resultingstatistically smoothed index. To illustrate, if the varianceof the predicted index is two-thirds the variance of theinitial estimate, the weight on the initial index, 0.4, istwo-thirds the weight on the predicted index, 0.6.Columns 3-8 of the table at the end of this documentpresent the initial, predicted, and statistically smoothedlog indices and their standard errors.Geographic Smoothing The law of one price implieswages in nearby counties cannot sustainably differ morethan justified by the cost of commuting between them.If the wage difference is larger, workers have anincentive to commute from the low wage county to thehigh wage county, increasing the supply of workers inthe latter and reducing it in the former, thereby reducingthe wage difference. Prior to adoption of the currentmethodology, neighboring counties sometimes hadimplausibly different FPLI values. Geographic smoothingensures index differences between nearby counties areconsistent with their proximity. To implementgeographic smoothing, the statistically smoothed indexvalue for each county is replaced by the higheststatistically smoothed value from a comparison group ofcounties, adjusted for the lost value commute time, ifthat value is higher.There were two broad problems with the marketbasket approach. First, it was subject to numerouschallenges to its accuracy. Second, not only was it at bestan indirect proxy for labor costs, but it systematicallymis-measured them. That is because, other things beingequal, places that are more productive, and thus moreattractive to firms, will have higher wages and prices,while places that are more pleasant in which to live, andthus more attractive to workers, will have lower wagesbut higher prices. Numerous independently publishedestimates of relative wage and price patterns imply thatthe market basket approach yields an index which is aless accurate reflection of relative labor costs thanmaking no adjustment at all. 4 Consequently, the currentcomparable wage approach unambiguously produces abetter measure of relative school personnel costs.The FPLI Calculation 5Initial Estimate The first step in calculating the FPLI is tomake an initial estimate of relative wage differencesbetween counties, holding occupation constant. Thismeans a county’s index is not impacted by having moreor less workers in high wage occupations, but rather byhaving higher or lower wages within given occupationscompared to the same occupations in other counties.Wage differences related to labor market size,measured by population or total employment, and dueto differences in land costs or commute times, are morepronounced for occupations that tend to locate atdenser locations within a labor market. The estimationprocedure controls for this tendency.Impact on School FundingFlorida adjusts state funding to provide all studentsaccess to substantially equal educational servicesappropriate to their needs. This involves equalization fordifferences in the value of the local property tax base per45Jim Dewey, (2005) Improvements to the 2003 Florida Price Level Index,available at .The data and Stata code for the 2020 FPLI calculation are available athttps://drive.google.com/file/d/1BeGlLUFf5kI CIGa0z35YCKanicvkDK/view?usp sharing.3

student and adjustment for differences in operatingcosts across districts. Indeed, the very factors that createdifferences in the property tax base per student alsocreate differences in the cost of education. 6Cost differences depend on differences in thequantity of inputs needed to meet the standard ofeducation and on the per unit cost of those inputs.Differences in the quantity of inputs needed arerepresented by elements of the funding calculation likeProgram Cost Factors, the ESE Guaranteed Allocation,the Sparsity Supplement, and the Class Size ReductionAllocation. The District Cost Differential (DCD) adjustsfor differences in the per unit cost of inputs. It assumeslabor makes up 80% of operating costs, relying on theFPLI to represent them, and that the other 20%, forexample textbooks, cost the same everywhere.The figure below illustrates the relative importanceof the DCD among the adjustments to school funding.The grey circular markers represent what funding wouldhave been if the state engaged in no resourceequalization. The flat line represents what fundingwould have been if all funds were allocated on an equalper student basis with no regard for cost differences.The vertical distance between unequalized funding andflat funding illustrates the largest effect of Florida’sfunding system—allocating more state funding tostudents in districts with less taxable value per student.The grey triangles indicate funding if the DCD wereeliminated but all else remained the same. Thedifference between funding with no DCD and flatfunding represents the combined impact of alladjustments other than the DCD. The squares indicateactual funding. The difference between actual fundingand funding with no DCD indicates the impact of theadjustment for labor costs. While the impact of the DCDis not negligible, for most districts it is tiny compared toequalization for differences in the tax base and smallerthan the impact of the other adjustments as well.Ongoing Study-Geographic SmoothingThe methodology has evolved over time to makeimprovements where possible and to adapt to changingcircumstances as needed. This section discusses work toimprove geographic smoothing. For the 2010 index,values in 23 counties containing 12.8% of the state’spopulation were replaced by commute cost adjustedvalues from another county in geographic smoothing.For the 2020 index, 41 counties containing 29.6% of thestate’s population were replaced. With the increase inthe share of the state’s population directly affected, theimpact on other counties through the state averagegrew as well.6For more detail on state and local school funding in Florida, see theFlorida Department of Education report 2020-21 Funding for FloridaSchool Districts, available t/Fefpdist.pdf.4