Document Sample
Roadmap Powered By Docstoc
					The Returns to Education in
  the Early 20th Century:
   New Historical Evidence

  Joseph Kaboski    Trevon D. Logan
 The Ohio State U   The Ohio State U
                       and NBER
I.    Overview
II.   The Historical Record
III. Our Model
IV. New Historical Data
V.    Empirical Results
VI. Robustness, Extensions and Implications
• There is an important recent literature on the
  interaction between skill, technology, and
  economic growth– such as the skill biased
  technical change literature.
• Beginning in the late 19th century, skill began to
  replace physical capital, raw materials, and
  unskilled labor, leading to high returns to
  education in the early 20th century.
• There is evidence that the returns to education
  over the 20th century were U shaped.
• We do not know how much regional variation
  there was in returns to education at the
  beginning of the 20th century.
• The historical record gives us some reasons to
  expect significant variation in the returns to
  – Segmented labor and capital markets
  – Different levels of industrial composition
  – Different natural resource endowments
• This paper looks at regional variation in the
  returns to education at the beginning of the
  20th century
                 Main Results
• We construct a two sector model which
  highlights the idea that differences in initial
  technological endowments will lead to
  differences in the returns to education.
• We exploit a new data source, a 1909 Report to
  the Commissioner of Education, to estimate the
  returns to education of high school teachers in the
  South, Midwest, and West.
• We find evidence of significant variation in the
  returns to education in the early 20th century.
        The Historical Record I
• There was marked regional heterogeneity in
  the factors (raw materials, physical capital,
  unskilled labor) that lead to high returns to
  education in the early 20th century
  – The South had a different capital market
  – Literacy rates varied by region
  – The extent of manufacturing varied by region
  – The capital intensity of agriculture varied by
  – The nature and type of natural resources varied by
        The Historical Record II
• There is also evidence that the tenure in
  manufacturing firms was longer than
  previously thought, which could possibly lead
  to further investments in technology.

• While there is a literature that looks at the
  development of these differences in technology
  and industrial development by region, by the
  beginning of the 20th century these differences
  were part of the technological/capital
     Supply and Demand Factors
• If the returns to education reflect the supply
  and demand for skill, we can think of the
  heterogeneity as giving us evidence that there
  was variation in supply and demand.
• Supply factors (literacy rate) relatively fixed
• Demand factors (extent of manufacturing,
  value of machinery, etc.)
• While we can intuit about returns to ed. in high
  demand/low supply and low demand/high
  supply scenarios, it is unclear what returns we
  should expect from low/low or high/high
Supply and Demand Factors in 1910
      Factor          Georgia   Wisc.   Texas   Calif.
     Literacy          76.7     96.3    87.8    95.5
  Man. % of LF*        64.1     96.5    91.2    50.8

Livestock Val. PC      31.0     67.9    81.8    53.6
Mach. Val. Per Cap.    8.0      22.7    14.6    15.4
Mach. Val. Per Ag.     25.7     119.4   48.1    105.3

      *Percent changes from 1900 to 1910
               The Model I
• We construct a two-sector model to generate
  predictions about the returns in education in
  the early 20th century.
• We assume that there are two sectors of
  production, a land/resource-intensive sector
  and a capital-intensive sector.
• Both sectors use skilled and unskilled labor
  and capital or land to produce output.
• Each region is a small open economy that
  takes the relative price of output as given.
               The Model II
• In the initial equilibrium, the fraction of
  workers employed in the capital-intensive
  sector is increasing in the capital/land ratio,
  and the relative wage of skilled workers is
  decreasing in their size.

• We then consider the introduction of a new
  capital dependent sector that is more skill
  dependent than the old capital-intensive sector.
         Predictions From the Model
•    The model yields four predictions:
    1) If the productivity of the new technology is sufficiently large,
       the new sector displaces the old capital-intensive sector, and
       the new technology employs a higher fraction of high skilled
       workers than low skilled workers.
    2) The number of high skilled employed in the new technology
       exceeds the number of high skilled employed in the old capital-
       intensive technology. The relative wage of high skilled workers
    3) The higher the capital/land ratio, the higher the fraction of
       skilled and unskilled workers employed in the new technology
       and the higher the relative wage of skilled workers.
    4) The new technology raises the return to capital relative to land.
       Furthermore, the higher the ratio of skilled/unskilled labor, the
       larger the increase in the relative rental rate of capital.
I.    Overview
II.   The Historical Record
III. Our Model
IV. New Historical Data
V.    Empirical Results
VI. Robustness, Extensions and Implications
                 The Data I
• We use a 1909 report by Edward Thorndike to
  estimate the returns to education of high
  school teachers.
• The report was the first in a five report plan to
  analyze secondary education in the U.S.
• This data is the earliest we know of that allows
  us to estimate the returns to education, and the
  earliest to do so by region.
• The data is culled from the summary reports of
  a survey given to approximately 5000 high
  school teachers in the US, chosen to be
  representative at the time.
                The Data II
• The data list the annual salary, education, and
  years of experience by sex for high school
  teachers in Ohio, Wisconsin, Illinois,
  California, Texas, and Georgia.
• The data for OH/IL/WI was grouped together
  because the responses were deemed similar.
• Thorndike sent out a supplementary survey to
  gauge the extent of measurement error and
  misreporting of education and experience; he
  found that the first survey did not suffer from
  large amounts of measurement error or
  aggregation bias. We use the first survey.
              Summary Statistics
               Georgia OH/IL/WI   Texas    Calif.
Avg. Salary     828      848      733      1142
                (377)    (379)    (278)    (316)

Avg.            12.6     12.6     12.6     13.8
Schooling       (1.7)    (1.9)     (1.9)    (1.4)

Avg.            8.2      9.1      9.6       8.3
Experience      (5.8)    (7.2)     (7.1)    (7.0)

Fraction        67%     57%       64%      34%
N               137      3141     381      658
          Empirical Results I
• We use this information to estimate the returns
  to education for teachers in Texas, California,
  Georgia and Ohio/Illinois/Wisconsin.
• The results show that there is marked
  geographical variation in the returns to
  education for high school teachers. Teachers
  in Georgia have lower returns than teachers in
  the Midwest.
• As expected, returns in Texas are high and
  returns in California are low.
             Empirical Results II
               Georgia OH/IL/WI     Texas     Calif.
Schooling       .033      .070      .071      .005
                (2.06)    (23.33)    (7.88)    (0.83)

Experience      .012      .048      .034      .034
                (0.92)     (12.0)    (4.86)    (8.5)

Experience^2   .0004      -.0008    -.0009    -.0007
                (0.66)     (8.0)     (3.0)     (7.0)

Male             .64       .16       .27       .22
                (10.67)    (16.0)    (9.0)     (11.0)

N               137       3141       381       658
         A Caveat – Median Data
• The data for OH/IL/WI is pooled, and it is not
  possible to generate state-specific estimates based on
  individual data for these states.
• We do have data on median incomes by sex,
  education and experience for each state (OH, IL, WI).
• We use individual data from CA, TX, & GA to create
  median data for those states to see how well it tracks
  with their individual returns.
• Since the overall pattern from the CA, TX, & GA
  regressions is consistent with their individual returns,
  we use the median returns to estimate the returns to
  education for each Midwestern state.
       Median Return Estimates
               Ohio      Illinois   Wisconsin
Schooling      .080       .073        .034
                (8.89)    (8.11)       (3.4)

Experience     .030       .026        .042
                (3.33)    (2.89)       (4.2)

Experience^2   -.0003    .0001       -.0006
                (1.0)     (0.33)       (2.0)

Male           .079       .192        .274
                (1.68)     (4.0)      (5.37)

N                133       122          99
            Generalizability I
• Unfortunately, we only have evidence from
  high school teachers. This raises the question
  of generalizability.
• Do the returns to education for teachers track
  with general returns to education overall?
• To answer this question we must turn to the
  present– we use the IPUMS 5% samples from
  the 1980, 1990, and 2000 census to see how,
  by state, the returns for teachers (all levels)
  track with overall returns to education.
            Generalizability II
• By state, we regress the overall return to
  education for each Census year on the return for
  teachers and time dummies.
• While we do find a positive relationship there
  are two important caveats
  – Teachers are not distinguished by type (primary,
    secondary) in the Census, so we must use the
    returns to all teachers.
  – The dramatic changes in the qualifications of
    teachers in the 20th century (and lack of data)
    doesn’t allow us to say with certainty whether the
    relationship between teacher returns and all returns
    was strong in the first half of the century.
                    Robustness I
•    We believe that these estimates reflect the
     returns to education and not simply salary
     variation for teachers
    1) Variation in teachers’ education levels was large.
    2) The estimates of returns could not be inferred from
       looking at average salary and schooling by state.
•    We also stratify the sample by gender and
     experience (>5 yrs. of experience)
    –   We divide the sample by gender because men had
        more outside options other than teaching
    –   We divide the sample by experience because those
        with fewer years of teaching as less “tied” to the
             Robusteness II
                 Men       Less Experience
Georgia          .020           .012
                  (0.95)        (0.4)

OH/IL/WI         .075           .052
                 (18.75)        (1.3)

Texas            .085           .071
                  (9.44)        (7.1)

California       .004           .020
                  (0.4)         (2.0)
• What does this finding of regional heterogenity
  tell us about returns to education from 1910 to
• The returns to education were high in the
  beginning of the 20th century, and teachers’
  returns to education do decline over time, and
  then begin to increase again post 1980.
• Also, the returns for teachers track well with
  the general pattern of overall returns.
                                        Figure 1: Comparison of Mincerian Returns over Time



Mincerian Return

                   0.08                                                                  Just Teachers, All States, Sex Dummy
                                                                                         All Workers, All States, Sex Dummy
                                                                                         Just Teachers, Samples States, Sex Dummy
                   0.06                                                                  All Workers, Samples States, Sex Dummy



                          1940   1950   1960     1970     1980      1990     2000
                                                 Year                                                          Source: IPUMS
• There was marked heterogeneity in the returns to
  education in the early 20th century.
• Regions that had a large capital endowment had
  higher returns to education.
• The returns that we estimate for high school
  teachers track well with overall returns, and the
  returns to teachers also show a U shaped pattern
  over the 20th century.
• Endowments matter for the returns to education,
  and depending on the endowment, regions may
  experience increasing or U-shaped returns to

Shared By: