Green Jobs: Who Benefits? Demographic Forecasting of Job Creation in U.S. Green Jobs Studies more

THE HARRIS SCHOOL | THE UNIVERSITY OF CHICAGO Green Jobs: Who Benefits? Demographic Forecasting of Job Creation in U.S. Green Jobs Studies Kyle B. Gracey, candidate MSESP ‘10 June 11, 2010 More than 20 studies have attempted to assess net job creation through the growth in green jobs (Kammen et al 2004, Green 2008). None have considered what the demographics of these job holders might be. Using 2000-2009 gender and race percentages from the Current Population Survey for detailed occupation and industry categories, a variety of periods of lagged linear regressions provide forecasts of the race, gender, and Latino and Hispanic ethnicity of these jobs through 2017. Many forecasts show poor statistical quality due to limited observations, especially with multi-period lags. Despite this, most come close to the employment patterns in the Department of Labor’s Employment Projections for 2018. Applying the forecasts to the categories of jobs considered in the existing green jobs studies, whites and males appear to occupy the majority of green jobs generated. This holds even if we assume an unrealistic, linear extrapolation of the percentage point growth in fractions of jobs held by women and minorities for the types of jobs most produced in these green jobs studies. However, if the green jobs studies are accurate, and even if the forecasts are biased by a dozen percentage points, overall women, blacks, Asians, and Hispanics and Latinos would still gain jobs in most studies considered, though blacks and Asians are relatively more susceptible to no gain in jobs if the percentage forecasts applied here are significantly far above the actual. This is one of three research endeavors in my exploration of the interdisciplinary nature of the M.S. Environmental Science and Policy degree. The second, on the eutrophication potential of phosphatebased fertilizer runoff in the Somali coastal Large Marine Ecosystem, is under development with Dr. Pamela Martin in the Geophysical Sciences Department and Dr. Gidon Eshel in the Physics Department of Bard College, and we aim to submit it for publication this summer. The third, on a methodology for the economic valuation of biodisparity, will be presented at the Society for Conservation Biology’s International Congress for Conservation Biology in July. Special thanks to Dr. Dan Black for early guidance on the development of a research project focused on the labor economics of U.S. green jobs. Thanks to Fay Booker for Stata troubleshooting. 2 Table of Contents The Universe of Green Jobs Studies and Demographics of Green Jobs .................................................... 3 Materials and Methods ............................................................................................................................... 5 Defining the Scope of Green Jobs Studies ................................................................................................ 2 Table 1: Studies Reviewed .................................................................................................................... 6 Demographic Data by Occupation and Industry ....................................................................................... 7 Table 2: Example of Current Population Survey Demographic Data .................................................... 8 Forecasting Demographics for Years After 2009 ...................................................................................... 9 Results ........................................................................................................................................................ 12 Demographics for Studies Using Past Years of Impact ........................................................................... 12 Table 3: Pollin and Wick-Lim 2008 ...................................................................................................... 12 Table 4: White and Walsh 2008 .......................................................................................................... 14 Table 5: Pollin et al 2008 ..................................................................................................................... 15 Graphic 1: A series of one-period lag linear regression forecasts for the percentage of the… ........ 16 Table 6: AIC Scores by Demographic and Industry/Occupation Database ......................................... 17 Graphic 2: Best AIC Score Histograms by Year 2010 .......................................................................... 20 Durbin h-test Summary ........................................................................................................................... 22 Linear Regression with Newey-White Errors .......................................................................................... 22 Table 7: Comparison of 2018 Employment Projections to 2017 Forecasts by occupation… ............ 22 Table 8: Pollin et al 2009… ................................................................................................................. 23 Table 9: Redefining 2009 .................................................................................................................... 36 Table 10: Booz 2009............................................................................................................................ 40 Table 11: DEFINING 2008 .................................................................................................................... 42 Discussion ................................................................................................................................................... 44 Green Jobs – Who Benefits? ................................................................................................................... 44 Uncertainty and Error ............................................................................................................................. 45 Further Research ........................................................................................................................................ 46 References .................................................................................................................................................. 48 Appendix A: Stata Estimation Code ........................................................................................................... 51 3 The Universe of Green Jobs Studies and Demographics of Green Jobs Since 1998, more than 20 (Kammen et al 2004, Green 2008) studies have attempted to estimate the U.S. employment impact of green jobs (a term discussed in more detail below). Some modeled changes in the number and types of workers in a variety of primarily private sector industry categories and occupation types from new government investments or other policies, particularly the introduction of carbon prices. Some compare these job outputs to that from equivalent investments in traditional outputs, particularly equivalent amounts of electricity produced from coal, oil, or natural gas plants compared to renewable energy (which typically excludes nuclear energy) electricity generation. Most attempt to forecast the impact of these investment or policy changes into the near future (between 2010 and 2030; see Table 1). Others simply compare the current number and occupational categories of workers in facilities producing components necessary to generate “green” products (especially renewable energy) to those employed producing similar, traditional products (again, often coal-, oil-, or natural gas-fueled electricity). Each study uses a different scope to define what jobs are “green”. Some do not use the term at all, instead discussing “clean energy” jobs (those employed in firms producing renewable energy-based electricity). All studies include these renewable energy-based jobs. Some also include positions in mass transit, building and/or automobile energy efficiency improvements, and/or biofuels. The most expansive consider impacts on all types of jobs in the Census Bureau’s Census Occupation Codes or all industries in their Census Industry Classification system (industry 2009, occupation 2009). In 2010, the U.S. Department of Labor announced its intent to begin defining and counting green jobs (Green 2010). Some studies also consider additional characteristics that these green jobs will have or that the jobs will require, most commonly education level (DEFINING 2008, White and Walsh 2008) and wage or household income (Pollin et al 2008, Pollin et al 2009). None, however, consider the gender or race of 4 these employees, while, for example, a recent projection of Recovery Act effects does briefly consider gender impacts (Romber and Bernstein 2009). While gender disparities in jobs producing renewable energy in developing countries have been reviewed widely (see, for example, Skutsch (2003) and Clancy et al (2004)), evaluations and forecasts of green job creation in the United States lack these demographic considerations. Knowledge of which demographics of people will likely benefit most from green jobs may impact the desirability of policies designed to promote green jobs, provide information on which types of green jobs will impact which demographics most, and provide information that may aid in the adoption or termination of policies that impact these job demographics within the context of larger green jobs efforts. Materials and Methods Defining the Scope of Green Jobs Studies Taking into consideration the variable definition of green jobs discussed above, and rather than employing a particular definition to limit what supposedly “green jobs” studies were considered, I attempted to locate any study produced in 2000 or later that evaluated job impacts of policies or industries designed to reduce environmental impacts and that covered jobs in the entire United States, including all of the studies listed in Kammen (2004) and Green (2008). From there, I narrowed the studies considered to those that calculated employment impacts for specific categories of jobs or specific industry classifications, since studies that only calculated total job creation in the United States would not provide sufficient detail to reliably calculate the race and gender of job holders, given that these characteristics vary greatly depending on job category or industry (Labor 2009; Highlights 2009). Table 1 summarizes the studies considered after narrowing, including the year in which the job creation is expected to occur (either forecasted or current at the time of publication). The Uncertainty section discusses the validity of these studies, including a review of some recent criticisms in the literature. 5 Table 1: Studies Reviewed Job/Industry Categories Title (Year of Publication) Included Select industries and occupations, some from DEFINING, ESTIMATING, AND FORECASTING THE Census Classification RENEWABLE ENERGY AND ENERGY EFFICIENCY system (all detailed INDUSTRIES IN THE U.S. AND IN COLORADO Standard Occupation (2008) Classifications for Colorado jobs) Select Census Occupation Classification for select GREEN JOBS STUDY (2008) Census Occupation Classification GREEN PROSPERITY: How Clean-Energy Policies Select Census Occupation Can Fight Poverty and Raise Living Standards in Classification the United States (2009) Green Recovery: A Program to Create Good Jobs and Start Building a Low-Carbon Economy (2008) GREENER PATHWAYS: Jobs and Workforce Select Census Occupation Development in the Clean Energy Economy Classification (2008) JOB OPPORTUNITIES FOR THE GREEN Select Census Occupation ECONOMY: A STATE-BYSTATE PICTURE OF Classification OCCUPATIONS THAT GAIN FROM GREEN 2007 2008 Select Census Occupation Classification 2008-2010 Unspecified post-2009 2013 2030 Year of Impacts 6 INVESTMENTS (2008) Redefining the Prospects for Sustainable Prosperity, Employment Expansion, and Summary Census Industry Environmental Quality in the US: An Assessment Classification of the Economic Impact of the Initiatives Comprising the Apollo Project (2003) program (~2013) Ten Years after start of Demographic Data by Occupation and Industry A comparable set of demographic data for these jobs is needed, ideally one that is based on the same Census Occupation and Industry Classification system. The Census Bureau’s American Community Survey provides yearly information on detailed industry and occupation from 1996 onward (A 2009). However, in 2006 the Census Bureau began including samples from individuals in group quarters1, and estimates comparing data from before and after this change are not recommended for any geographic scope where significant group quarters exist (Ibid). The Bureau of Labor Statistics Current Population Survey provides an alternative. Although from a smaller sample size than the American Community Survey (60,000 households monthly versus about 1.3 million people yearly) (Ibid, Labor 2004), it provides the same national, annual2 demographic data for the same Census Industry and Occupation Classifications from 1995 forward. The classification systems were revised in 2003 to use the 2002 Census Occupation and Industry Classifications, and only corrections from 2000 onward have been produced (Ilg 2010). I use published data on race, gender, and Hispanic or Latino identity from the Current Population Survey by detailed Census Occupation “A [Group Quarter] facility is a place where people live or stay that is normally owned or managed by an entity or organization providing housing and/or services for the residents.” (Ibid) 2 1 Although demographic data is also available on a monthly basis over the same time period, it is not available at the level of detailed Census Occupation or Industry Classification. Using monthly data would provide more observations, but would also require longer out-of-sample forecasts. 7 Classification (occupation 2000-2009) and detailed Census Industry Classification (industry 2000-2009) for 2003 through 2009, and unpublished versions of the same tables for 2000 through 2002 provided by Bureau of Labor Statistics economists (Ilg 2010, Bowler 2010). This data provides a ten-year time series (ten observations) for each of 535 occupation and 317 industry categories for total number of workers and percentages of female, black, Asian, and Hispanic or Latino identity workers in each category and year. The exception is Asian percentages of job holders by detailed occupation, for which data is only available from 2003 through 2009. Gender, race, and Hispanic or Latino identity are overlapping categories for individual workers, and data on, for example, race by gender is not available. Table 2 provides an example of the dataset. Despite the examples in Table 2, not all occupation or industries have 10 years of data. Some years contain no data for unreported reasons, or if the number of workers nationally is below 50,000. Between 7 and 13 occupations and industries report no data for these reasons in each demographic considered, were dropped from consideration, and are not a part of the industry and occupation totals above. Table 2: Example of Current Population Survey Demographic Data Percentage of Hispanic or Latino employees by occupation Year Chief executives General and operations managers 2000 2001 2002 3.1 2.3 2.8 4.2 6.3 6.9 2003 3.3 7.6 2004 3.7 7.1 2005 3.8 6.2 2006 2007 2008 2009 4.6 5 4.8 4.6 7.7 7.9 6.2 6 For the studies evaluating green jobs holders in previous years, applying that year of data to the categories of jobs considered shows which demographics of job holders may be affected in each study. Tables 3 and 4 review these estimates. For considering projections of workers in future years, it is necessary to estimate the demographics for each job or occupation category in these years. 8 Forecasting Demographics for Years After 2009 To compute their own forecasts of expected green jobs, all of the studies projecting future years of green jobs rely largely on input-output tables to estimate the sectoral impacts of investments or policy changes, specifically the impacts of changes in one sector on all other sectors. All of the studies use the Impact Analysis for Planning (IMPLAN) (Pollin and Wicks-Lim 2008) input-output software. Without access to this software, or precise information on how each study’s estimates were constructed (none are published in the peer-reviewed literature, though most include detailed but not comprehensive explanations of their models and assumptions), other strategies become necessary to attempt to forecast job demographics. The limited number of observations for each demographic characteristic severely constrains the choice of forecasting models. Any multivariate regression that attempts to include other predictive variables in each time period would further reduce already limited degrees of freedom. Following the method of Diebold (2007), I consider autoregressive models of the form ∑ which yields model errors of the form ∑ where n [1..7], t [2010..2017], and ∑ has White Noise distribution WN(0,σ2)3. Graphic 1 shows an (1) (2) example of regressions with this model. Attempting to predict values beyond 2017 using this model would result in regressions with F-distributions with zero degrees of freedom. For models where more than one period of lag is considered, the maximum forecastable year is further reduced by one year for This assumption, though common for time series, is unverifiable since Stata does not permit Bartlett’s or Portmanteau white noise tests for multiple panel IDs, such as in these demographic datasets. 3 9 each additional lag period considered. Thus, eight successive years of one-period lags again yields a regression with an F-distribution with zero degrees of freedom and is not considered here. All regressions were performed in Stata/SE 10.1 (StataCorp 2009). Appendix A provides Stata code for all regressions, forecasts, and postestimation tests discussed in this study. Despite the loss in observations and degrees of freedom from autoregressive processes of more than order one, the additional lag periods may provide better estimates of the forecasted years. Comparisons of the Akaike Information Criterion (AIC) across models of varying numbers of lagged regressors suggests which model provides the lowest forecast risk by assuming all models are approximations of the true process and values extra lags if they improve the forecast but penalizes additional lags or other regressors4. This is preferable in forecasting to the use instead of the Bayes or Schwarz Information Criterion, which assumes the existence of a true model (Armstrong 2001). Table 5 presents a summary of AIC scores by autoregressive process. Graphic 2 presents a histogram of which autoregressive process had the lowest /best AIC score for each demographic and occupation/industry. Comparing the latest year of available forecasted values (2017) for each occupation code against the same occupations projected for 2018 in the Bureau of Labor Statistics Employment Projection table (which does not provide employment projections for detailed industry/occupation by gender/race/ethnicity, but does provide detailed occupation projections for total workers in the economy) allows an additional check of the forecasting model’s accuracy. The Employment Projection 4 Stata’s AIC command assumes that the number of observations used for estimation (T) and the number of observations (N) in total are always equal, which is not true for forecasts. Instead, the AIC is calculated directly by using the stored regression values: AIC = ln(_result(4)/_result(1))*10+(1+_result(3))*2 where _result(4) returns the residual sum of squares, _result(1) returns T, _result(3) returns the autoregression period, and 10 is replaced by the appropriate N (help 2009). 10 estimates are based on multivariable projections of factors that are believed to influence each occupation category’s growth or decline (Employment 2009). Table 7 compares the two projections. Durbin’s h-test is employed to check the assumption of serially uncorrelated errors (Armstrong 2001). Results are summarized below. Regressions with Newey-West standard errors are also performed in addition to the original regressions with robust standard errors in case the error terms are serially correlated in addition to being heteroskedastic. Regressions with Newey-West standard errors in Stata do not compute root mean squared error, which is necessary for calculating confidence bands around the forecasts (see Graphic 1), so while the Newey-West regressions are used to examine the model, the robust standard error linear regressions are used for the out-of-sample forecasting (Hansen 2010). For the 2030 study (DEFINING 2008), however, not only is this 13 years outside of the predictive period, but the likelihood of cyclical impacts on demographic employment rates becomes more likely, especially if recessions occur in that period (Rives and Sosin 2002, Clark and Summer 1981). Calculating a simple linear extrapolation for 2017 through 2030, using both the observed and predicted data, while not likely a reliable predictor of actual demographic employment data, provides perhaps an extreme bound on the fraction of jobs held by each demographic. This still provides an interesting comparison of how much or little these demographics might occupy in the projected green jobs. Graphic 1 includes an example of this extrapolation. Results Demographics for studies using past years of impact Tables 3 and 4 present demographics for studies estimating jobs in previous years (the same years in which the studies were prepared). If no percentage was available, either the immediately prior 11 or next year was used, with preference given to following years when available. -- indicate that fewer than 50,000 workers of that demographic held the job or an estimate was otherwise unavailable in the dataset. Table 3: JOB OPPORTUNITIES FOR THE GREEN ECONOMY: A STATE-BYSTATE PICTURE OF OCCUPATIONS THAT GAIN FROM GREEN INVESTMENTS (Pollin and Wicks-Lim 2008) 2007 % Female % Black % Asian % His/Lat 1.7 5.9 1.2 14.3 0.9 6.3 2.2 13.2 (2008) 1.9 5.6 1.8 26.9 0.9 4.9 0.1 42.9 (2008) 1.9 4.5 0.6 (2008) 5.9 23 1.5 26.7 (2008) 8.1 2.6 1.5 9.2 11.5 2.9 8.8 6.1 ---19.5 1.7 5.9 1.2 14.3 5.6 7.3 2.8 5.7 28.9 12.4 4.8 24.4 (2008) (2008) ---20 51.6 26.8 1.1 12.2 (2008) 55.1 11.6 1 12 20.8 4.9 29.4 2.8 8.6 6.9 13.5 4.1 22.4 8.3 5.9 11.2 5.6 7.3 2.8 5.7 28.9 12.4 4.8 24.4 (2008) (2008) ---20 ----0.9 6.2 0.3 8.8 3.7 4.8 2.2 11.8 (2008) “Green Economy Investment” “Representative Job” Electrician Heating/Air Conditioning Installer Carpenter Roofer Building Retrofitting Insulation Worker Industrial Truck Driver Construction Manager Civil Engineer Rail Track Layer Electrician Welder Metal Fabricator Engine Assembler Bus Driver Dispatcher Computer Software Engineer Electrical Engineer Engineering Technician Welder Metal Fabricator Engine Assembler Environmental Engineer Millwright Sheet Metal Worker Mass Transit Energy-Efficient Automobile Wind Power 12 Machinist Electrical Equipment Assembler Industrial Truck Driver Industrial Production Manager First-Line Production Supervisor Electrical Engineer Electrician Industrial Machinery Mechanic Welder Metal Fabricator Electrical Equipment Assembler Laborer Construction Manager Chemical Engineer Chemist Chemical Equipment Operator Cellulosic Biofuels Chemical Technician Industrial Truck Driver Agricultural Inspector % of Total Workforce in 2007 -- 5.2 57.9 5.9 16.7 19.4 8.6 1.7 3.2 5.6 28.9 57.9 18.6 8.1 21.2 40.8 15.4 (2008) 32.4 5.9 -11.0 % Black 5 13.1 23 4.7 11.6 6.9 5.9 8.3 7.3 12.4 (2008) 13.1 17.9 2.6 10.3 6.8 16.2 (2008) 7.4 23 -- 5.2 16.7 1.5 3.7 4.8 13.5 1.2 2.6 2.8 4.8 16.7 1.9 1.5 11.6 18.3 4.4 (2008) 6 1.5 Solar Power 46.4 % Female 4.7 % Asian 12.1 (2008) 19.4 (2008) 26.7 (2008) 9.8 14.9 (2008) 4.1 14.3 11.3 5.7 24.4 (2008) 19.4 (2008) 30 9.2 4.3 5.1 13.2 (2008) 14.5 26.7 (2008) -14.0 % His/Lat The study in Table 3 looks at what types of jobs will benefit most from increases in clean energy production and increased energy efficiency efforts. It does not consider the number of workers who might be lost in other professions as a result of shifts in energy production or decreases, or slower growth in energy production from increased energy efficiency. In Table 3, Hispanics/Latinos are a minority in every job category, never higher than 42.9% and usually much lower. Asians and Blacks are even lower than this. The bottom row lists the fraction of the total workforce occupied by members of each demographic, according to the Current Population Survey. 13 Table 4: GREENER PATHWAYS: Jobs and Workforce Development in the Clean Energy Economy (White and Walsh 2008) 2008 % Asian 1.9 1.2 0.6 (2007) 0.1 2.2 -1.3 0.6 2.7 -2.4 4 -3.7 3.7 3.7 4.2 2.4 2.4 2.8 1.5 4.4 6.9 -3.8 4.8 % Asian Construction Laborer Sheet Metal Worker Insulation Worker Cement Mason & Concrete Finisher “ENERGY EFFICIENCY JOBS AT-A-GLANCE” Heating, air conditioning & refrigeration mechanic and installer Hazardous materials removal worker Carpenter Plumber, Pipefitter, & Steamfitter Electrician Boilermaker Laborers & freight, stock & material movers; hand Cutting, punching, and press machine setters, operators & tenders Drilling, boring, & machine tool setters, operators & tenders Customer service representative Welders, cutters, solderers, & blazers Production, planning, & expediting clerks Machinist Maintenance & repair workers; general Laborers & freight, stock & material movers; hand Shipping, receiving, & traffic clerks Truck drivers; heavy & tractor-trailer Chemical equipment operators & tenders Chemical Technician Electrical & electronic equipment repairers, commercial & industrial Sales representatives, wholesale & manufacturing, technical…. % of Total Workforce in 2008 % Female 3.1 4.8 1.9 (2007) 0.6 (2009) 2 -1.5 1.4 1 -17.1 20.2 -68.3 4.7 58.2 6.9 3.5 17.1 32.8 8.9 15.4 35.2 -27.3 46.7 % Female % Black 7.7 6.2 4.5 (2007) 7.4 8.6 -6 6.4 5.9 -15.9 9.1 -18.3 8.7 8.9 7 10.2 15.9 11.6 23.4 16.2 18 -3.9 11.0 % Black % His/Lat 44.1 11.8 -57.7 13.2 -25.7 19.5 16.2 -21.2 22 22 (2007) 14.5 21 7.3 12.1 13.7 21.2 20.2 26.7 13.2 7 -8.6 14.0 % His/Lat “WIND JOBS AT-A-GLANCE” “BIOFUELS JOBS AT-AGLANCE” 14 The study from Table 4 also only looks at categories of job holders currently in select renewable energy fields. It does not provide data on how many people work in the renewable energy industry. Again, Hispanics and Latinos tend to have larger shares of the jobs considered compared to their overall fraction of the workforce, while the opposite is true for women. The bottom row again lists the fraction of the total workforce occupied by members of each demographic, according to the Current Population Survey. Table 5: Green Recovery: A Program to Create Good Jobs and Start Building a Low-Carbon Economy (Pollin et al 2008) This study uses the same categories and clean energy groupings as in Table 4, except that here Smart Grid replaces Energy Efficient Automobile above. All other results are nearly identical and are omitted. 2009 % % Black Asian 5.3 26.6 5.1 17 11.6 13.2 5.1 5.7 7.4 1.9 5.5 0.8 10 0.7 10.7 4.7 % % Black Asian "Green Economy Investment" "Representative Job" Computer Software Engineer Electrical Engineer Electrical Equipment Assembler Machinist Construction Laborer Operating Engineer Electrical Power Line Installer/Repair Smart Grid % of Total Workforce in 2009 % Female 20.2 9.4 59.4 5.4 2.7 1.5 1.3 47.3 % Female % His/Lat 3.5 5.1 28.8 15 44.2 13.7 10.8 14.0 % His/Lat For most categories, women and blacks occupy a smaller fraction of the jobs than they do overall jobs in the economy. The opposite is true for Asians, and Hispanics and Latinos have as many job categories below the overall percentage in the economy as they do above. Again, though, each job category is not weighted for the number of jobs held in each job category. 15 Graphic 1: A series of one-period lag linear regression forecasts for the percentage of blacks holding occupations coded as “Chief Executives” % Black Chief Executives 5 4 3 2 1 2000 2010 Year Percent 99% Forecast Interval 90% Forecast Interval 95% Forecast Interval Extrapolation Forecast 99% Forecast Interval 90% Forecast Interval 95% Forecast Interval 2020 2030 Graphic 1 shows an example of a the data produced from each linear regression forecast with robust standard error calculations, using one-period lags and forecast ranges using the root mean squared error, as well as the extrapolation calculation. Extrapolations were bottom- or top-coded at 0% or 100%, respectively. The t-statistics on many of the lagged regressors across the 100,000+ regressions calculated are low (approximately 20% of variables and constants with associated p<0.1). However, tstatistics are less useful in forecasting since the percentages of the demographics in each job category may not change (that is, they may be truly be zero) over some years and yet this would still provide a forecast of future percentages if these values were true. Still, with so few observations, the low t-statistics are troubling, especially given their even higher frequency in job categories with much fewer than 10 observation years. Serially correlated errors 16 would further affect the forecasted values. Testing for correlation and regressions to account for this are discussed later in this section. In 2017, the errors converge to the forecasted value because the 7th single-period lagged linear regression has so few observations that it returns a root mean squared error of 0. Expectedly, this occurs for all job categories with sufficient observations to forecast this far. Table 6: AIC Scores by Demographic and Industry/Occupation Database AIC_x is the AIC value for x autoregression periods. Variable AIC_1 1011 AIC_2 461 AIC_3 2 Obs -4.043436 -5.321989 -14.5881 Mean Std. Dev. 9.07499 10.04734 6.51269 Asian, Industry Min -46.05707 -57.85435 -19.19327 Max 21.33263 18.61589 -9.982933 Variable AIC_1 1661 AIC_2 1064 AIC_3 524 AIC_4 170 Obs 1.987925 .8188832 .5104573 -1.28286 Mean Std. Dev. 11.6139 13.19493 12.65508 13.76325 Black, Industry Min -66.66119 -124.0927 -74.28119 -39.16174 Max 37.32703 33.55389 28.52877 26.41997 Variable Obs Mean Std. Dev. Min Max 17 AIC_1 1660 AIC_2 1067 AIC_3 529 AIC_4 172 3.645581 2.332412 1.607696 -2.042613 11.41918 12.36967 13.09098 15.20435 -47.72913 -54.10381 -69.10397 -59.64277 35.80178 33.98428 29.12156 27.4439 Hispanic/Latino, Industry Variable AIC_1 1665 AIC_2 1066 AIC_3 529 AIC_4 172 Obs 5.720758 4.578044 3.661514 1.091639 Mean Std. Dev. Min Max 49.59357 47.02958 43.93935 39.47191 13.10844 -83.47218 13.29101 -56.9099 13.22128 -47.9027 14.08396 -37.84316 Women, Industry Variable AIC_1 2251 AIC_2 1484 AIC_3 860 AIC_4 281 Obs -1.140123 -1.520837 -3.006933 Mean Std. Dev. Min Max 33.84206 32.73219 30.50871 26.6098 11.23221 -97.11179 11.21848 -43.44164 12.51348 -60.59088 16.14255 -82.95504 -6.953628 Asian, Occupation Variable Obs Mean Std. Dev. Min Max 18 AIC_1 2335 AIC_2 1539 AIC_3 892 AIC_4 294 4.389343 3.699485 2.293809 -.4355545 11.22309 12.05429 13.41656 16.76334 Black, Occupation -66.11866 -124.8761 -70.51173 -89.03619 41.97369 39.82377 37.45194 34.13867 Variable AIC_1 2305 AIC_2 1523 AIC_3 829 AIC_4 274 Obs 6.500856 5.794316 3.53945 .9987584 Mean Std. Dev. Min Max 45.96317 43.39228 41.05086 38.80769 11.99973 -84.02988 12.05895 -60.19807 13.87875 -60.95013 16.42046 -74.73439 Hispanic/Latino, Occupation Variable AIC_1 2328 AIC_2 1542 AIC_3 899 AIC_4 294 Obs 6.829033 6.332378 4.758618 1.225794 Mean Std. Dev. 12.92031 12.8418 14.15988 16.94505 Female, Occupation Min -96.16654 -37.44624 -69.96892 -67.00991 Max 60.57936 56.87597 52.64983 46.71107 19 The AIC scores tend to show that the most lagged periods tend to yield better model predictions, though not always. A single-period lagged regression also fares well across categories, with the advantage of allowing more prediction years. Graphic 2 shows a set of histograms for the same scores for models predicting year 2010, indicating for how many occupations or industries for that demographic the AIC score indicates the particular number of lagged periods is the best model for predicting the 2010 fraction of employees. In general, though, none of the AIC scores are very high, and many are negative. Graphic 2: Best AIC Score Histograms for Year 2010 150 Frequency 100 Frequency 1 2 best 3 4 0 1 50 0 50 100 150 2 best 3 4 Female, Occupation Black, Occupation 150 Frequency 100 50 0 1 2 best 3 4 0 1 20 40 Frequency 60 80 100 2 best 3 4 Latino/Hispanic, Occupation Asian, Occupation 20 100 80 Frequency 40 60 0 20 1 2 best 3 4 0 1 20 Frequency 40 60 80 2 best 3 4 Female, Industry Black, Industry 60 Frequency 40 Frequency 1 2 best 3 4 0 1 50 0 20 100 150 80 1.5 2 best 2.5 3 Hispanic/Latino, Industry Asian, Industry 21 Durbin h-test Summary The mean Durbin alternative test for autocorrelation varied greatly among the regressions estimated (range of Prob > chi2 ~[0.304..0.986]) and with a mean ~0.531, suggesting that autocorrelation is likely present in most model specifications, and Newey-White standard error calculations would yield more accurate error bounds (though, again, not feasible for calculating the forecast intervals). Liner Regression with Newey-White errors F-statistics are noticeably higher compared to the equivalent regression with robust standard error calculations. Overall, approximately 10 percentage points more of the total number of regressions show variables and constants with p<0.1, or at least improved t-statistics, compared to the robust standard error calculations. This also suggests serial correlation of errors. Table 7: Comparison of 2018 Employment Projections to 2017 Forecasts by occupation category The table presents a sample of the comparisons between 2018 projected data (left) and 2017 forecasted data (right) for total number of employees in the category. Overall, forecasted data exceeds projected data by about 11% across categories, but this averaged percentage is not weighted for the number of employees in each category. If the projected data is drawn from a better model (it is at least a more complex model) than the forecasted data from this study, it suggests that the forecasted data used here may, on average, overestimate the fractions of employment by the target demographics. BLS Projected (2018) Accountants and auditors Actors Actuaries Advertising and promotions managers Aerospace engineers Agents and business managers of artists, performers, and athletes Agricultural and food science technicians Agricultural and food scientists 1,570.0 63.7 23.9 43.9 79.1 27.8 23.8 35.9 Study Forecasted (2017) Accountants and auditors 1733.636353 Actors 32.666668 Actuaries 27.666666 Advertising and promotions managers 77.5 Aerospace engineers 132.166672 Agents and business managers of artists, performers, and athletes 43.333332 Agricultural and food science technicians 48 Agricultural and food scientists 32.5 Difference 10% -49% 16% 77% 67% 56% 102% -9% 22 Table 8: GREEN PROSPERITY: How Clean-Energy Policies Can Fight Poverty and Raise Living Standards in the United States (Pollin et al 2009) This study considers the impact of $150 billion in stimulus money directed toward clean energy or fossil fuels, finding that, overall, clean energy spending generates more jobs, but the job creation is (unsurprisingly) distributed differently among industries between the two types of spending (first three columns). As the year is unspecified but meant to occur shortly after 2009, comparisons are made to the 2010 industry category predictions, using the lagged model prediction with the highest AIC score in each industry category (next four columns). Values below zero have been bottom-coded. 90% confidence intervals are in parentheses (some may appear slightly unbalanced due to rounding). Most confidence intervals are in the range of a few percentage points above or below the mean value, though a few categories should very large confidence intervals (e.g. – more than 15 percentage points above and below the mean for Hispanics/Latinos in the oil and gas extraction industry). Values in italics and without confidence intervals are extrapolated values. -- indicates no data available. Blackened rows indicate aggregated or partial categories. These removals account for 44.41% of clean energy industry shares and 36.91% of fossil fuel industry shares. Calculations of demographic employment fractions were not made for these, since it is not clear from the study exactly which jobs from the original industry categories were included. The final four columns show the demographic percentages weighted by the share of that industry for the clean energy spending and fossil fuel spending, with a summary of the weighted percentages at the end. Women, Asians, and blacks appear to occupy a larger share of the overall workforce if the stimulus were invested in fossil fuels compared to clean energy, while Hispanics/Latinos (and, implicitly, whites and men) gain more from investments in clean energy than from fossil fuels. However, given that about 40% of the industries are not calculated, the results for the full economy may vary significantly (for example, the “passenger and ground transportation” aggregate industry occupies almost an 8 percentage point larger share in the clean energy investment scenario, and about 8 more percentage points have been excluded from the clean energy industries than from the fossil fuel industries). 23 2010 Forecast (90% Confidence Interval) TABLE A1. INDUSTRY EMPLOYMENT SHARES BY ENERGY SECTOR Industry Industry share of cleanenergy sector Industry share of fossil fuels sector Clean Fossil Demographic % * Industry Share/100 Clean Fossil Clean Fossil Clean Fossil Women farms forestry, fishing, and related activities 3.88% 1.03% 1.10% 0.10% Black Asian His/Lat Women Black Asian Hispanic/Latino mining oil and gas extraction 0.28% 15.77% 23.6 (18.328.9) mining, except oil and gas 0.27% 1.55% 6.0 (3.5- 3.3 (2.08.4) 4.7) 0 (01.1) 1.2 (04.1) 13.0 (0.030.3) 14.4 (9.918.9) 21.3 (18.224.4) 0.066 3.7217 0.0168 0.9462 0.0092 0.5204 0.036 2.0501 25.2 (20.430.1) 14.1 (12.515.7) 0.068 0.3906 0 0 0.0032 0.0186 0.039 0.2232 support activities for mining 0.01% 0.25% 6.1 (4.6- 1.7 (0.47.7) 3.1) 10.1 (8.911.4) 2.0 (1.32.8) 0.001 0.0353 0.0006 0.0153 0.0002 0.0043 0.002 0.05325 utilities 0.26% 2.27% 19.7 (17.222.2) 9.7 (8.111.2) 0.051 0.4472 0.0263 0.2293 0.0052 0.0454 0.025 0.22019 24 construction 25.69% 7.68% 9.8 (9.5- 5.5 (4.9- 1.6 (1.59.9) 6.0 1.7) wood products 0.81% 0.21% 22.6 (21.224.1) 14.1 (12.315.9) 17.5 (15.419.7) 2.518 0.7526 1.413 0.4224 0.411 0.1229 5.806 1.73568 manufacturing: durable goods 17.5 (14.619.2) 19.3 (17.521.1) 9.5 (8.2- 1.5 (0.610.8) 2.3) 0.142 0.0368 0.077 0.02 0.0122 0.0032 0.114 0.02961 nonmetallic mineral products 0.61% 0.23% 9.4 (6.3- 1.9 (1.412.5) 2.5) 0.118 0.0444 0.0573 0.0216 0.0116 0.0044 0.107 0.04025 primary metals fabricated metal products machinery computer and electronic products 0.47% 2.17% 0.38% 1.65% 1.66% 1.59% 0.63% 0.18% 33.4 (31.934.9) 0.85% 0.20% 16.6 5.9 (4.7- (15.77.1) 17.6) 9.1 (7.510.7) 0.531 0.0601 0.0938 0.0106 0.2639 0.0299 0.145 0.01638 electrical equipment, appliances, and components 31.6 (25.637.6) 11.9 (6.117.6) 6.1 (4.57.8) 9.8 (7.811.8) 0.269 0.0632 0.1012 0.0238 0.0519 0.0122 0.083 0.0196 25 motor vehicles, bodies and trailers, and parts 0.28% 0.31% other transportation equipment 0.04% 0.03% 41.5 -furniture and related products 0.23% 0.09% -- 51.1 0.017 0.0125 -18.8 (16.421.2) 16.5 (13.819.3) -- -- -- 0.02 0.01533 26.3 (22.829.7) 0.15% 0.19% 6.3 (2.2- 3.4 (1.510.5) 5.3) 0.06 0.0237 0.0145 0.0057 0.0078 0.0031 0.043 0.01692 miscellaneous manufacturing 34.8 (16.753.0) 0.48% 0.50% 7.8 (6.2- 4.7 (09.4) 10.1) 0.052 0.0661 0.0117 0.0148 0.0071 0.0089 0.025 0.03135 manufacturing: nondurable goods food and beverage and tobacco products textile mills and textile product mills 0.09% 0.08% 51.0 (50.551.5) 6.6 (2.4- 4.7 (1.710.8) 7.7) 12.2 (025.7) 0.046 0.0408 0.0059 0.0053 0.0042 0.0038 0.011 0.00976 26 apparel and leather and allied products 0.07% 0.08% paper products 0.18% 0.24% 40.8 (32.449.3) 14.6 (7.921.3) 3.7 (07.1) 48.7 (47.350.1) 0.073 0.0979 0.0263 0.035 0.0067 0.0089 0.088 0.11688 printing and related support activities 0.20% 0.28% 33.5 (31.835.1) petroleum and coal products 0.14% 2.19% 8.2 (6.4- 4.4 (2.89.9) 5.9) 17.1 (13.320.9) 11.4 (10.512.3) 10.5 (9.911.2) 19.9 (18.821.0) 12.4 (9.615.2) 10.1 (8.212.1) 14.7 (13.116.2) 0.067 0.0938 0.0164 0.2296 0.0088 0.0123 0.04 0.05572 20.7 (18.323.0) 0.37% 1.17% 3.8 (1.46.1) 6.5 (5.47.7) 0.029 0.4533 0.0239 0.3745 0.0053 0.0832 0.017 0.27156 chemical products 34.3 (33.035.6) 28.6 (26.031.2) 0.127 0.4013 0.0422 0.1334 0.0241 0.0761 0.037 0.11817 plastics and rubber products 0.81% 0.51% 3.1 (1.84.3) 0.232 0.1459 0.0851 0.0536 0.0251 0.0158 0.119 0.07497 wholesale trade 2.65% 3.55% 29.0 (28.229.8) 15.3 5.8 (4.5- 4.3 (4.2- (14.67.1) 4.4) 16.0) retail trade 0.769 1.0295 0.1537 0.2059 0.114 0.1527 0.405 0.54315 27 motor vehicle and parts dealers 0.79% 0.63% food and beverage stores general merchandise stores 1.05% 0.85% 1.07% 0.87% other retail 3.72% 3.00% transportation and warehousing air transportation 0.14% rail transportation 0.42% 0.17% 40.5 (37.543.5) 0.23% 13.0 (11.014.9) 6.8 (5.48.2) 1.5 (0.82.2) 8.5 (7.212.5) 0.057 0.0689 0.0182 0.0221 0.0095 0.0116 0.012 0.01445 8.6 (6.610.7) 0.028 0.0152 0.0508 0.0278 0.0063 0.0035 0.036 0.01978 4.8 (0.19.6) 0.005 17.0 (16.117.8) 1E03 0.00192 12.1 6.6 (3.4- (10.79.8) 13.5) water transportation truck transportation 0.02% 1.20% 0.04% 25.0 (15.334.7) 1.40% 12.9 (11.214.6) 9.7 (6.5- 4.8 (1.112.9) 8.5) 13.4 (12.014.8) 1.1 (0.71.5) 0.01 0.0019 0.0039 0.001 0.0019 0.155 0.1806 0.1608 0.1876 0.0132 0.0154 0.204 0.238 28 transit and ground passenger transportation 8.80% 0.25% pipeline transportation other transportation and support activities 0.02% 0.96% 10.6 5 -0 0.002 0.1018 0.001 0.048 --0 0 0.62% 0.83% warehousing and storage 0.33% 0.33% 28.9 (25.532.3) 0.37% 22.7 (19.226.2) 24.9 3.1 (1.9- (22.44.3) 27.3) 0.095 information 0.095 0.0749 0.0749 0.0102 0.0102 0.082 0.082 publishing industries (includes software) 0.29% 29 motion picture and sound recording industries 0.15% 0.18% broadcasting and telecommunications 0.67% 0.86% information and data processing services 0.12% 0.15% finance and insurance federal reserve banks, credit intermediation, and related activities 1.32% 1.96% 30 insurance carriers and related activities 1.02% 1.15% 61.9 (60.063.7) funds, trusts, securities, commodity contracts, and other investments 1.87% 2.50% 9.7 (8.6- 3.6 (3.110.7) 4.1) 8.7 (8.19.4) 0.631 0.7119 0.0989 0.1116 0.0367 0.0414 0.089 0.10005 real estate and rental and leasing, and other financial vehicles real estate 2.70% 3.48% 49.1 (47.151.0) 8.3 (7.4- 3.8 (2.89.2) 4.8) 12.1 (11.013.2) 1.326 1.7087 0.2241 0.2888 0.1026 0.1322 0.327 0.42108 31 rental and leasing services and lessors of intangible assets 0.50% 1.02% 25.5 (23.327.7) 12.5 14.2 (11.14.7 (3.2- (13.713.9) 6.2) 14.6) 0.128 0.2601 0.0625 0.1275 0.0235 0.0479 0.071 0.14484 professional, scientific, and technical services legal services 0.90% computer systems design and related 0.11% 1.49% 56.5 (54.758.4) 0.16% 7.2 (6.2- 2.7 (2.28.2) 3.2) 7.9 (6.88.9) 0.509 0.8419 0.0648 0.1073 0.0243 0.0402 0.071 0.11771 25.9 (24.027.8) miscellaneous professional, scientific, and technical services 6.00% 7.64% 17.3 6.3 (5.2- (15.87.4) 18.9) 4.9 (4.35.6) 0.028 0.0414 0.0069 0.0101 0.019 0.0277 0.005 0.00784 32 management of companies and enterprises 1.04% 3.22% 73.2 (66.180.3) 8.5 (4.5- 4.9 (0.9- 7.4 (6.412.5) 8.8) 8.5) 0.761 2.357 administrative and waste management services 0.119 0.2737 0.051 0.1578 0.077 0.23828 administrative and support services 4.82% 7.27% waste management and remediation services 0.18% 0.22% 15.4 (13.816.9) educational services 0.75% 0.76% 74.9 (74.475.3) 1.85% 13.8 (12.115.6) 11.0 (8.513.5) 1.4 (0.12.7) 12.2 (9.415.0) 0.028 0.0339 0.0248 0.0304 0.0025 0.0031 0.022 0.02684 3.7 (3.4- 9.1 ( 3.9) 8.4-9.8) 0.562 0.5692 0.0825 0.0836 0.0278 0.0281 0.068 0.06916 health care and social assistance ambulatory health care services 1.85% 33 hospitals 1.28% 1.28% 77.2 (75.878.7) 16.0 (15.016.9) 6.5 (6.26.7) 7.5 (6.38.8) 0.988 0.9882 0.2048 0.2048 0.0832 0.0832 0.96 0.096 nursing and residential care facilities 0.94% 0.94% social assistance 0.96% 0.96% 85.5 (84.886.2) 0.69% 19.7 (18.421.1) 13.8 3.5 (2.8- (12.54.3) 15.1) 0.821 0.821 0.1891 0.1891 0.0336 0.0336 0.132 arts, entertainment, and recreation 0.132 performing arts, spectator sports, museums, and related activities 0.58% amusements, gambling, and recreation industries 0.54% 0.59% accommodation and food services 34 accommodation 0.51% 0.63% 57.3 (55.759.0) 14.7 (13.316.1) 7.7 (6.29.3) 24.4 (22.626.1) 0.292 0.361 0.075 0.0926 0.0393 0.0485 0.124 0.15372 food services and drinking places 3.23% 3.80% 52.2 (51.652.8) 10.7 (10.211.2) 21.5 (20.123.0) 5.9 (5.66.2) 1.686 1.9836 0.3456 0.4066 0.1906 0.2242 0.694 0.817 other services, except government 4.26% 4.83% 51.6 (51.152.1) 10.2 (9.510.8) 18.6 5.9 (5.5- (17.96.3) 19.4) 2.198 2.4923 0.4345 0.4927 0.2513 government 0.285 0.792 0.89838 federal state and local Total 0.09% 0.40% 100% 0.12% 0.61% 99.71% 15.53 21.558 4.4059 5.5298 1.897 2.3213 10.93 9.22112 Clean Fossil Clean Fossil Clean Fossil Clean Fossil Women Black Asian Hispanic/Latino 35 Table 9: Redefining the Prospects for Sustainable Prosperity, Employment Expansion, and Environmental Quality in the US: An Assessment of the Economic Impact of the Initiatives Comprising the Apollo Project (2003) This study presents the detailed industry tables used in the preparation of the seminal Apollo Alliance green jobs study (New 2004). It models approximately $300 billion worth of Federal investments in a variety of renewable energy, energy efficiency, and greenhouse gas emission-reducing activities over 10 years of economic impacts. It estimates that, by the end of the ten year period, the initial investment more than pays for itself in government receipts, while generating 1.4 million jobs. 128 tables are presented for impacts in detailed industry sectors for each of the investment areas considered, as well as summary tables. Only the summary table (C 32) is evaluated here, although sufficient data exists to compare each table in the same manner. Assuming 2013 as the end of the 10-year period, 2013 forecasted demographic percentages are compared to the projected number of total jobs from the study to estimate how many jobs would be generated for each demographic. The formatting for the forecasted percentages and the choice of forecast model for each industry category follow that in Table 8. Revenue generation columns from the original table have been removed for clarity. In this case, differences in industry aggregation led to ~33% of the projected jobs not being evaluated for their forecasted demographic division. Data from the “New Construction” and “Maintenance & Repair Construction” is combined and compared against the “Construction” column in the forecast data since these two categories cover the construction industry in the same way that the “Construction” category does. A direct comparison is possible. For comparison, the final column shows the forecasted 2013 fraction that each demographic will occupy in the overall economy. The results indicate that, at least for the industry categories available for study, women will gain 30% of the new jobs estimated to be created, blacks 6%, Asians 4%, and Hispanics and Latinos 11%, though these fractions are smaller than the total fraction of jobs forecasted to be held by women, Asians, and Hispanics and Latinos in 2013. 36 "Table C.32 The Total Ongoing Annual Economic Impact of Apollo Project Initiatives (as of Year 10): Detailed Sectoral Results" % of Labor Force Employment (Permanent Jobs) Women 94,351 909 0 5.2 (3.1-7.4) 27.5 (21.51,205 33.6) 0 77,553 9.58 (9.49.8) 48,844 7,546 60.3 (59.25,989 61.5) 9,046 38.6 (34.67,294 42.6) 11,180 6,523 10,350 6,838 8,320 16.8 (15.314,272 18.3) 7.4 (6.2-8.5) 2.1 (2.02.3) 0 (0-1.0) 2013 Forecast Employment (Permanent Jobs) Sector Agricultural Products & Services Forestry & Fishery Products Coal Mining Crude Petroleum & Natural Gas Miscellaneous Mining New Construction Maintenance & Repair Construction Food Products & Tobacco Textile Mill Products Apparel Paper & Allied Products Printing & Publishing Chemicals & Petroleum Refining Rubber & Leather Products Lumber Products & Furniture Stone, Clay, & Glass Products Primary Metal Black Asian Lat/His Women Black Asian Lat/His 0 (0-1.0) 6.6 (2.6-10.6) -2.2 (2.02.5) 1.7 (1.61.7) 7.1 (3.510.8) 1.7 (0.23.2) 9.2 (3.215.1) 27.1 (24.829.4) 21.8 (13.430.2) 331.38 79.53 26.51 110.86 5.0 (4.6-5.4) 12,108.83 6,319.85 2,148.75 34,253.59 3.1 (0-8.4) 3,611.37 185.66 425.22 1,305.60 22.7 (12.31.9 (0-4.2) 33.1) 2,815.48 - 138.59 1,655.74 14.0 (12.515.6) 2,397.70 1,056.13 299.71 1,998.08 37 Fabricated Metal Products Machinery, Except Electrical Electric & Electronic Equipment Motor Vehicles & Equipment Transp. Equip., Exc. Motor Vechicles Instruments & Related Products Miscellaneous Manufacturing Transportation Communication Electric, Gas, Water, Sanitary Services Wholesale Trade Retail Trade Finance Insurance Real Estate Hotels, Lodging Places, Amusements 23,388 15,661 16,646 27.7 (26.613,394 28.7) 26.9 (25.29,313 28.6) 7,077 37.8 (35.9135 39.6) 45,267 5,831 1,209 150,896 319,123 23,572 16,644 13,690 21,078 71.8 (70.110,177 73.4) 160,875 66.5 (66.111.6 (11.112.0) 15.4 (13.913.1 (13.113.2) 3.5 (1.614.9 (13.716.2) 11.5 (11.422.6 28.9 (25.532.2) 29.3 (28.729.9) 48.9 (48.649.3) 52.1 (50.953.2) 61.9 (60.563.4) 48.3 (45.651.0) 7.8 8.3 (7.0-9.6) 7.4 (6.1-8.7) 9.7 (9.4-10.1) 8.7 (7.7-9.8) 10.7 (9.212.1) 9.1 (8.4-9.9) 12.1 12.2 3.5 (2.97.1 (2.04.1) 12.3) 15.6 (14.94.2 16.2) 12.5 (11.77.9 13.3) 4.8 (4.610.1 (9.74.9) 10.4) 3.7 (3.39.4 (8.64.0) 10.3) 3.5 (2.212.5 (11.53.8) 13.5) 1,317.81 349.40 44,212.53 156,051.15 12,281.01 10,302.64 6,612.27 454.82 100.35 11,166.30 30,954.93 2,050.76 1,780.91 705.55 42.32 6,337.63 25,210.72 1,131.46 615.83 711.38 85.84 23,539.78 39,890.38 2,380.77 1,564.54 1,711.25 6.0 (5.5-6.5) 7.6 (5.99.2) 16.6 (11.921.3) 51.03 8.10 10.26 22.41 13.1 (12.014.3) 11.9 (10.013.9) 9.4 (8.67.2 10.2) 6.3 (6.38.4 (6.96.3) 9.9) 3,710.14 2,505.20 1,754.61 1,108.25 964.37 586.72 1,259.04 782.29 13690*.091 479.15 Personal Services Business Services 7,307.09 1,180.53 24,774.75 1,325.33 1,516.37 38 Eating & Drinking Places Health Services Miscellaneous Services Households Total (excluding grayed rows) % of Total Jobs Above % of Total Workforce in 2013 66.9) 52.2 (51.857,185 52.7) 71,690 90,423 92.1 (90.88,923 93.4) 1,392,415 16.8) 10.7 (10.211.2) 5.3) 6.3 (5.96.7) 11.6) 23.1 (21.824.4) 106,981.88 29,850.57 6,118.80 5,630.63 3,602.66 18,500.63 13,209.74 6.4 (3.2-9.6) 1.4 (1.21.6) 42.4 (38.646.1) 8,218.08 571.07 124.92 3,783.35 411,016 89,665 49,806 148,282 30% 6% 4% 11% 45% 11% 4% 15% Women Black Asian Lat/His 39 Table 10: GREEN JOBS STUDY (Booz 2009) This study calculates the number of jobs created by the growth in the demand for Leadership in Energy and Environmental Design (LEED) buildings for the period 2000-20008, and forecasts the same employment and economic data for the period 2009-2013. It predicts net job creation in 5 industry categories most related to green buildings at 7.9 million by 2013 from LEED building construction alone, though with a net loss of jobs in some industry categories. Employment impacts in each forecast year are not provided, so the estimate below instead applies the 2013 forecasted demographic percentages to the total 2009-2013 job creation estimated. Thus, the actual (even assuming the forecasts are accurate) should vary a small amount due to differences in the demographic percentages across the years 2009-2012 compared to 2013. The same formatting and data treatment methods as in the previous table apply here. Coverage is 99.87% of jobs affected since the Water, Sewage, and Other Systems category was removed due to differences in aggregation compared with the CPS industry categories. Of the total jobs predicted to be created, women occupy 10% of them, blacks 5%, Asians 3%, and Hispanics and Latinos 27%. Latinos and Hispanics appear to gain more jobs in LEED-related occupations compared to their overall projected share of jobs in 2013, as do blacks (though by a much slimmer margin). Women, especially, and Asians are forecasted to occupy a smaller fraction of the LEED-related jobs than in the U.S. economy overall. 40 “Impact of Green Construction Spending by NAICS Industries” Nonresidential Construction Residential Construction Electric Power Generation, Transmission, and Distribution Water, Sewage, and Other Systems Waste Management Remediation Services Total % of Total Jobs Created % of Total Workforce in 2013 20092013 Women 7,497,566 9.6 (9.49.8) 461,443 24.0 (19.1(41,745) 28.9) (10,401) 15.2 (12.9(4,398) 17.6) 7,902,466 1.0 (0.21.8) 12.9 (11.214.6) % of Labor Force Asian Black 3.4 (3.25.0 (4.6-5.4) 3.6) 1.5 (1.41.6) 9.0 (7.2-10.7) 2013 Forecast Hispanic 27.1 (24.829.4) 19.3 (16.122.6) Women 764,064.86 (10,018.80) # in Labor Force Asian Black 270,606.31 (626.18) 397,950.45 (3,757.05) Hispanic 2,156,891.44 (8,056.79) 24.2 (23.325.2) (668.50) (43.98) (567.34) (1,064.32) 753,377.57 269,936.15 393,626.06 2,147,770.34 10% 3% 5% 27% 45% 11% 4% 15% 41 Table 11: DEFINING, ESTIMATING, AND FORECASTING THE RENEWABLE ENERGY AND ENERGY EFFICIENCY INDUSTRIES IN THE U.S. AND IN COLORADO (2008) The study estimates the change in select jobs for three scenarios above the baseline assumption of renewable energy industry growth. The table below contains the job creation assumptions for the “advanced case” of both aggressive technological improvements and government policies and spending. The extrapolated values for 2030 are used to calculate the number of employees in each demographic holding these estimated jobs. Again, the extrapolated values can at best be thought of like extreme values of where the demographics could develop, not reliable forecasts. The same formatting and other data treatment approaches from previous tables are repeated. Employment numbers for each job category are estimated from a bar graph and may be +/- 1,000. Interestingly, even in this extreme example, the demographics considered still occupy minority shares, overall, of the jobs estimated to be created, at least for the categories of jobs shown (but only barely for Hispanics and Latinos). Women hold 30% of the jobs, blacks 21%, Asians, 18%, and Latinos and Hispanics 49%. 42 Figure VII-2 U.S. Jobs Created By Renewable Energy In 2030 Compared to 2007 (Total Jobs Created -- Selected Occupations) Accountant Chemist Computer Systems Analyst Construction Manager Cost Estimor Electrician Environmental Scientist Iron & Steel Worker IT Manager Mechanical Engineer Management Analyst Power Line Worker Sheet Metal Worker Welder Total % of Total 95,000 22,000 38,000 28,000 28,000 90,000 21,000 10,000 19,000 38,000 41,000 23,000 21,000 41,000 515,000 2030 Extrapolated Women 91.3 0.0 6.9 0.0 46.5 6.8 31.5 100.0 0.0 10.0 39.3 0.0 47.0 0.0 Black 0.0 100.0 0.0 0.0 2.4 41.2 0.0 51.9 0.0 0.0 69.7 8.7 29.9 15.9 Asian 12.1 0.0 40.8 0.0 13.4 54.5 1.4 0.0 7.5 9.0 9.1 16.3 0.0 1.0 Hispanic/Latino Women 22,000 672 37,080 5,190 28,577 2,001 6,279 6,519 154,895 30% 108,318 21% Black Asian 11,495 15,504 3,752 49,050 294 1,425 3,420 3,731 3,749 410 92,830 18% Hispanic/Latino 95,000 16,852 19,456 8,064 17,010 19,000 39,852 8,924 26,527 250,685 49% 100.0 86,735 76.6 51.2 2,622 28.8 0.0 13,020 18.9 6,120 0.0 6,615 0.0 10,000 100.0 0.0 3,800 97.2 16,113 38.8 0.0 9,870 64.7 - 43 Discussion Green Jobs – Who Benefits? The forecasts and evaluations from this study can be viewed in two ways. First, women, Asians, blacks, and those of Hispanic or Latino ethnicity are forecasted to hold a minority of the green jobs that these studies indicate have been or will be produced, with whites and males occupying the majority. In that sense, these demographics may be expected to benefit the least from green job growth or programs that promote green jobs. Second, however, even if women, Asian, black, and Hispanic and Latino workers will constitute minorities of green jobs holders, that does not mean they will not benefit at all from a growth in green jobs. All occupy minority shares of the total workforce today (see bottom of Table 5), so it is not surprising that they would occupy minority shares in the near future, especially for those demographics that are far below 50%. Even if the forecasted fractions in future years are biased by a dozen percentage points, in some cases, these demographics still show net increases in jobs. For some studies, though, the margin for error is much smaller. All of the green jobs studies that calculate a net change in jobs in the U.S. estimate that total jobs available will grow with a shift toward more green jobs. This is largely based on calculations that green industries, particularly renewable energy electricity generation (and energy efficiency savings that divert the need for some increased energy production) are more labor-intensive than their nonrenewable alternatives. Pollin et al (2008) estimate that 1.5 million more jobs would be created from $100 billion of spending on their “green recovery program” than on the same amount of spending on the oil industry or household consumption. Unless all of those additional jobs go toward whites and males, which is unlikely given that these other demographics considered here make up at least some share of most of the job types and industries considered, all demographics of workers considered here 44 will likely gain jobs in total, provided that the underlying studies are accurate (more sources of estimation uncertainty and error are discussed below), compared to the counterfactual cases in which energy efficiency and renewable energy investments, supportive policies, and expected sector growth do not occur. Many studies also do not provide information on job categories created through induced or indirect5 jobs that are generated as a result of the creation of the direct jobs. These jobs, everything from accountants to housekeepers, may be smaller in number than the direct jobs created, but may more favorably employ the demographics of workers left out of the majority of direct green jobs. Most studies considered here model aggregate indirect and induced jobs. Uncertainty and Error Several sources of error likely reduce the estimation accuracy of the forecasted demographics. First, the Current Population Survey exhibits sampling and nonsampling errors that bias the previous year percentages (Employment 2006). The forecast model adds omitted variable bias since it does not assume to be the exact, true model, and many factors may affect labor force participation by these demographics in particular industries and occupations (Reskin et al 1999), none of which are considered here. The trends that have driven changes in demographic participation in 2000-2009, and that thus implicitly drive the forecasted years, may vary in coming years (for example, a possible slowdown in the growth of Hispanic and Latino fractional occupation), biasing the forecasts. The limited number of observations disrupts the model fit to the data. If the errors are serially correlated, the use of linear regressions with robust rather than Newey-White error calculations may further misrepresent the errors. Attempts to account for and introduce random, compounding errors for each year of forecasting Indirect jobs are those that support the direct green jobs but do not necessarily perform functions directly related to the provision or construction of the “green” product or service. This includes support staff such as accountants, secretaries, and lawyers. Induced jobs are those created by the other job holders spending the money they earn at their job. These jobs could include things like housekeepers or store owners (Pollin et al 2009). 5 45 produced an unknown computation error in Stata that I was unable to resolve, so the forecast confidence intervals in years after 2009 are likely too narrow. It is not certain that the same demographics of workers who occupy these occupation and industry categories overall will occupy them in green jobs with the same frequency. Perhaps female chemical engineers will choose more biofuel jobs while male chemical engineers will seek out petroleum companies. We do not know. These forecasts treat each change in the demographics of employment as independent. In reality, changes in the employment rates in some categories of jobs will likely affect the rate at which other jobs are filled by certain demographics. Even if we did understand the patterns of green job employment, we would have to assume that the studies themselves are accurate in their overall job estimates. A few studies have criticized the methodologies of some of the green jobs research considered here (Green 2008, Morris et al 2009). These criticisms, like most of the studies considered, have also not been published in peer-reviewed journals, though some or all of their criticisms may be valid nonetheless. Further Research At the minimum, future green jobs studies could include demographic data in their calculations, whether they are reports about current green jobs or estimates of future green jobs scenarios. A more sophisticated modeling of demographic changes from these green jobs policies would allow for more reliable demographic forecasts, particularly if the demographics could be incorporated into the existing input-output modeling that the studies (or future studies) use to derive their data. Future estimates might also want to incorporate additional demographics, such as worker age, or race and ethnicity by gender, to further subdivide the types of workers who might be expected to hold green jobs. 46 Understanding if certain demographics show noticeably varying rates of green job adoption versus non-green job adoption in the otherwise same occupation or industry category also seems important. Research might also consider what policies, such as targeted job training or recruitment programs, could shift the demographics of green jobs workers, and whether such shifts would be desirable. Do different demographics of workers show higher or lower average performance in green jobs, especially compared to equivalent non-green jobs, and if so, why? All of these questions and more remain unanswered, even as interest in green jobs continues. 47 References Help bic_note. in Help Contents. 2009. StataCorp LP. College Station, TX. Green jobs. in U.S. Department of Labor, Bureau of Labor Statistics [database online]. Washington, D.C., 2010 [cited 04/01 2010]. Available from http://www.bls.gov/green/. Jobs impact of a national renewable electricity standard. 2010. Navigant Consulting, . A compass for understanding and using American Community Survey data: What PUMS data users need to know. 2009. Washington, D.C.: U.S. Department of Labor, Housing and Household Economic Statistics Division, U.S. Census Bureau, . Highlights of women's earnings in 2008. 2009. Washington, D.C.: U.S. Department of Labor, Bureau of Labor Statistics, July 2009 (accessed September 20, 2009). Industry and occupation. in U.S. Census Bureau, Housing and Household Economic Statistics Division [database online]. Washington, D.C., 2009 [cited 01/05 2010]. Available from http://www.census.gov/hhes/www/ioindex/view.html. Labor force characteristics by race and ethnicity, 2008. 2009. Washington, D.C.: U.S. Department of Labor, Bureau of Labor Statistics, 1020. American Community Survey General Handbook. 2008. Washington, D.C.: U.S. Department of Labor, Housing and Household Economic Statistics Division, U.S. Census Bureau, , http://www.census.gov/acs/www/Downloads/ACSGeneralHandbook.pdf (accessed April 10, 2010). DEFINING, ESTIMATING, AND FORECASTING THE RENEWABLE ENERGY AND ENERGY EFFICIENCY INDUSTRIES IN THE U.S AND IN COLORADO. 2008. The American Solar Energy Society & Management Information Services, Inc., (accessed October 2, 2009). Green Jobs: A Review of Recent Studies. 2008. Austin: Center for Energy Economics, Bureau of Economic Geology, The University of Texas at Austin, http://www.beg.utexas.edu/energyecon/documents/CEE_Green_Jobs_Review.pdf (accessed October 20, 2009). Labor force statistics from the current population survey overview. in U.S. Department of Labor, Division of Labor Force Statistics, Bureau of Labor Statistics [database online]. Washington, D.C., 2004 [cited 11/02 2009]. Available from http://www.bls.gov/cps/cps_over.htm (accessed March 4, 2010). New Energy for America: The Apollo Jobs Report: For Good Jobs & Energy Independence. 2004. The Institute for America’s Future & The Center on Wisconsin Strategy, http://apolloalliance.org/downloads/resources_ApolloReport_022404_122748.pdf (accessed September 15, 2009). Redefining the prospects for sustainable prosperity, employment expansion, and environmental quality in the US: An assessment of the economic impact of the initiatives comprising the apollo project. 2003. Waco, TX: The Perryman Group, . StataCorp LP. Stata S/E 10.1. 2009. College Station, TX. Current population survey, household data annual averages, employed persons by detailed industry, sex, race, and Hispanic or Latino ethnicity. 2000-2009. Washington, D.C.: U.S. Department of 48 Labor, Division of Labor Force Statistics, Bureau of Labor Statistics, 18, http://www.bls.gov/cps/tables.htm (accessed January 10, 2010). Current population survey, household data annual averages, employed persons by detailed occupation, sex, race, and Hispanic or Latino ethnicity. 2000-2009. Washington, D.C.: U.S. Department of Labor, Division of Labor Force Statistics, Bureau of Labor Statistics, 11, http://www.bls.gov/cps/tables.htm (accessed January 10, 2010). Employment and Earnings. 2006. Washington, D.C.: U.S. Department of Labor, Division of Labor Force Statistics, Bureau of Labor Statistics, http://www.bls.gov/cps/eetech_methods.pdf (accessed January 11, 2010). Employment by occupation, 2008 and projected 2018. 2009. Washington, D.C.: U.S. Department of Labor, Division of Labor Force Statistics, Bureau of Labor Statistics, ftp://ftp.bls.gov/pub/special.requests/ep/ind-occ.matrix/occupation.xls (accessed December 20, 2009). Armstrong, J. S. 2001. Principles of forecasting: A handbook for researchers and practitioners. Kluwer Academic Publishers. Booz Allen Hamilton. 2009. Green jobs study. Washington, D.C.: U.S. Green Building Council, (accessed January 10, 2010). Bowler, Mary. 2010. RE: [cps] 2000-2002 CPS table 18. April 5, 2010. Clancy, J., S. Oparaocha, and U. Roehr. 2004. Gender equity and renewable energies. Thematic Background Paper 12, . Clancy, J. S., M. Skutsch, and S. Batchelor. 2003. The gender-energy-poverty nexus: Finding the energy to address gender concerns in development. DFID Project CNTR998521. Clark, Kim B., and Lawrence H. Summers. 1981. Demographic differences in cyclical employment variation. The Journal of Human Resources 16, (1) (Winter): 61-79. Diebold, F. X., R. J. Barro, J. E. Marthinsen, R. Carbaugh, B. V. Yarbrough, R. M. Yarbrough, P. Mourdoukoutas, B. Kaufman, J. L. Hotchkiss, and T. Hyclak. 2004. Elements of forecasting. Thomson. Hansen, Bruce. 2010. Distribution of Least‐Squares. Madison, WI: University of Wisconsin-Madison. http://www.ssc.wisc.edu/~bhansen/390/390Lecture16.pdf (accessed February 15, 2020). Ilg, Randy E. 2010. RE: [cps] multi-year CPS table 11?. February 16, 2010 (accessed April 30, 2010). Kammen, D. M., K. Kapadia, and M. Fripp. 2004. Putting renewables to work: How many jobs can the clean energy industry generate. RAEL Report, Renewable and Appropriate Energy Laboratory, University of California, Berkeley: 4-5. Laitner, Skip, Stephen Bernow, and John DeCicco. 1998. Employment and other macroeconomic benefits of an innovation-led climate strategy for the united states. Energy Policy 26, (5) (4): 425-32. Nogee, A., J. Deyette, and S. Clemmer. 2007. The projected impacts of a national renewable portfolio standard. The Electricity Journal 20, (4): 33-47. 49 Morriss, Andrew P., William T. Bogart, Andrew Dorchak, and Roger E. Meiners. 2009. Green jobs myths. SSRN eLibrary. Pollin, R., H. Garrett-Peltier, J. Heintz, and H. Scharber. 2008. Green recovery: A program to create good jobs and start building a low-carbon economy. Center for American Progress, Washington DC. Pollin, R., J. Wicks-Lim, and H. Garrett-Peltier. 2009. Green prosperity: How clean-energy policies can fight poverty and raise living standards in the united states. Published Studies. Pollin, Robert, and Jeannette Wicks-Lim. 2008. JOB OPPORTUNITIES FOR THE GREEN ECONOMY: A STATE-BY-STATE PICTURE OF OCCUPATIONS THAT GAIN FROM GREEN INVESTMENTS. Amherst, MA: Political Economy Research Institute, University of Massachusetts, Amherst, . Reskin, B. F., D. B. McBrier, and J. A. Kmec. 1999. The determinants and consequences of workplace sex and race composition. Annual Review of Sociology 25, (1): 335-61 (accessed March 10, 2010). Rives, J. M., and K. Sosin. 2002. Occupations and the cyclical behavior of gender unemployment rates. Journal of Socio-Economics 31, (3): 287-99 (accessed October 15, 2010). Romer, C., and J. Bernstein. 2009. The job impact of the American Recovery and Reinvestment Plan. January 8, : 2009. Skutsch, M. 2003. Tooling up for gender in energy. Unpublished Document. White, S., and J. Walsh. 2008. Greener pathways: Jobs and workforce development in the clean energy economy. Center on Wisconsin Strategy. 50 Appendix A: Stata Estimation Code Note: Data has already been formatted and set as time series by panel ID (occupation or industry title), and IDs with no observations have been removed. There are slight coding differences, not shown, for some demographics based on varying numbers of job categories and fewer years of observations for Asians sorted by industry. tsappend, gen float forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen add(21) AIC = . number = 1/7 { float AIC_`number' = . number = 0/7 { float y201`number' = . number = 0/6 { float y201`number'_2 = . number = 0/5 { float y201`number'_3 = . number = 0/4 { float y201`number'_4 = . number = 0/3 { float y201`number'_5 = . number = 0/2 { float y201`number'_6 = . number = 0/1 { float y201`number'_7 = . number = 0/7 { float y201`number'U = . number = 0/6 { float y201`number'U_2 = . number = 0/5 { float y201`number'U_3 = . number = 0/4 { float y201`number'U_4 = . number = 0/3 { float y201`number'U_5 = . number = 0/2 { float y201`number'U_6 = . 51 } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen } forvalues gen number = 0/1 { float y201`number'U_7 = . number = 0/7 { float y201`number'L = . number = 0/6 { float y201`number'L_2 = . number = 0/5 { float y201`number'L_3 = . number = 0/4 { float y201`number'L_4 = . number = 0/3 { float y201`number'L_5 = . number = 0/2 { float y201`number'L_6 = . number = 0/1 { float y201`number'L_7 = . number = 0/7 { float y201`number'_90_U = . number = 0/6 { float y201`number'_90_U_2 = . number = 0/5 { float y201`number'_90_U_3 = . number = 0/4 { float y201`number'_90_U_4 = . number = 0/3 { float y201`number'_90_U_5 = . number = 0/2 { float y201`number'_90_U_6 = . number = 0/1 { float y201`number'_90_U_7 = . number = 0/7 { float y201`number'_90_L = . number = 0/6 { float y201`number'_90_L_2 = . number = 0/5 { float y201`number'_90_L_3 = . number = 0/4 { float y201`number'_90_L_4 = . 52 } forvalues number = 0/3 { gen float y201`number'_90_L_5 = . } forvalues number = 0/2 { gen float y201`number'_90_L_6 = . } forvalues number = 0/1 { gen float y201`number'_90_L_7 = . } forvalues number = 0/7 { gen float y201`number'_99_U = . } forvalues number = 0/6 { gen float y201`number'_99_U_2 = . } forvalues number = 0/5 { gen float y201`number'_99_U_3 = . } forvalues number = 0/4 { gen float y201`number'_99_U_4 = . } forvalues number = 0/3 { gen float y201`number'_99_U_5 = . } forvalues number = 0/2 { gen float y201`number'_99_U_6 = . } forvalues number = 0/1 { gen float y201`number'_99_U_7 = . } forvalues number = 0/7 { gen float y201`number'_99_L = . } forvalues number = 0/6 { gen float y201`number'_99_L_2 = . } forvalues number = 0/5 { gen float y201`number'_99_L_3 = . } forvalues number = 0/4 { gen float y201`number'_99_L_4 = . } forvalues number = 0/3 { gen float y201`number'_99_L_5 = . } forvalues number = 0/2 { gen float y201`number'_99_L_6 = . } forvalues number = 0/1 { gen float y201`number'_99_L_7 = . } gen Percent_1 = Percent set more off forvalues number = 1(10)5351 { capture noisily reg Percent if ID==`number',r 53 replace AIC = ln(_result(4)/_result(1))*10+(1+_result(3))*2 if ID==`number' & Year==2010 *** One Period Lag *** capture noisily reg Percent L.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*9+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' *** Can't newey for first period lag forecast because Stata won't accept the pur trend equation when newey-ing by ID *** replace y2010=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2010 replace y2010U=temp+1.96*e(rmse) if ID==`number' replace y2010L=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp } capture noisily reg Percent L2.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*8+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2011 replace y2011U=temp+1.96*e(rmse) if ID==`number' replace y2011L=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L2.Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L3.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*7+(1+_result(3))*2 if ID==`number' & Year==2012 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2012=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2012 replace y2012U=temp+1.96*e(rmse) if ID==`number' replace y2012L=temp-1.96*e(rmse) if ID==`number' replace y2012_90_U=temp+1.645*e(rmse) if ID==`number' replace y2012_90_L=temp-1.645*e(rmse) if ID==`number' replace y2012_99_U=temp+2.576*e(rmse) if ID==`number' replace y2012_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L3.Percent if ID==`number', lag(3) 54 estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L4.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*6+(1+_result(3))*2 if ID==`number' & Year==2013 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2013=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2013 replace y2013U=temp+1.96*e(rmse) if ID==`number' replace y2013L=temp-1.96*e(rmse) if ID==`number' replace y2013_90_U=temp+1.645*e(rmse) if ID==`number' replace y2013_90_L=temp-1.645*e(rmse) if ID==`number' replace y2013_99_U=temp+2.576*e(rmse) if ID==`number' replace y2013_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L4.Percent if ID==`number', lag(4) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L5.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*5+(1+_result(3))*2 if ID==`number' & Year==2014 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2014=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2014 replace y2014U=temp+1.96*e(rmse) if ID==`number' replace y2014L=temp-1.96*e(rmse) if ID==`number' replace y2014_90_U=temp+1.645*e(rmse) if ID==`number' replace y2014_90_L=temp-1.645*e(rmse) if ID==`number' replace y2014_99_U=temp+2.576*e(rmse) if ID==`number' replace y2014_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L5.Percent if ID==`number', lag(5) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L6.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*4+(1+_result(3))*2 if ID==`number' & Year==2015 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2015=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2015 replace y2015U=temp+1.96*e(rmse) if ID==`number' replace y2015L=temp-1.96*e(rmse) if ID==`number' replace y2015_90_U=temp+1.645*e(rmse) if ID==`number' replace y2015_90_L=temp-1.645*e(rmse) if ID==`number' replace y2015_99_U=temp+2.576*e(rmse) if ID==`number' replace y2015_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L6.Percent if ID==`number', lag(6) 55 estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L7.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2016 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2016=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2016 replace y2016U=temp+1.96*e(rmse) if ID==`number' replace y2016L=temp-1.96*e(rmse) if ID==`number' replace y2016_90_U=temp+1.645*e(rmse) if ID==`number' replace y2016_90_L=temp-1.645*e(rmse) if ID==`number' replace y2016_99_U=temp+2.576*e(rmse) if ID==`number' replace y2016_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L7.Percent if ID==`number', lag(7) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L8.Percent if ID==`number',r if _rc==0 { replace AIC_1 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2017 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2017=temp if ID==`number' replace Percent_1=temp if ID==`number' & Year==2017 replace y2017U=temp+1.96*e(rmse) if ID==`number' replace y2017L=temp-1.96*e(rmse) if ID==`number' replace y2017_90_U=temp+1.645*e(rmse) if ID==`number' replace y2017_90_L=temp-1.645*e(rmse) if ID==`number' replace y2017_99_U=temp+2.576*e(rmse) if ID==`number' replace y2017_99_L=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L8.Percent if ID==`number', lag(8) estimates table ., star(.1 .05 .01) drop temp } *** Two Period Lag *** capture noisily reg Percent L(1/2).Percent if ID==`number',r if _rc==0 { replace AIC_2=ln(_result(4)/_result(1))*8+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2010_2=temp if ID==`number' replace y2010U_2=temp+1.96*e(rmse) if ID==`number' replace y2010L_2=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L_2=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp 56 } capture noisily reg Percent L(2/3).Percent if ID==`number',r if _rc==0 { replace AIC_2 = ln(_result(4)/_result(1))*7+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011_2=temp if ID==`number' replace y2011U_2=temp+1.96*e(rmse) if ID==`number' replace y2011L_2=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L_2=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(2/3).Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(3/4).Percent if ID==`number',r if _rc==0 { replace AIC_2 = ln(_result(4)/_result(1))*6+(1+_result(3))*2 if ID==`number' & Year==2012 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2012_2=temp if ID==`number' replace y2012U_2=temp+1.96*e(rmse) if ID==`number' replace y2012L_2=temp-1.96*e(rmse) if ID==`number' replace y2012_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2012_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2012_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2012_99_L_2=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(3/4).Percent if ID==`number', lag(3) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(4/5).Percent if ID==`number',r if _rc==0 { replace AIC_2 = ln(_result(4)/_result(1))*5+(1+_result(3))*2 if ID==`number' & Year==2013 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2013_2=temp if ID==`number' replace y2013U_2=temp+1.96*e(rmse) if ID==`number' replace y2013L_2=temp-1.96*e(rmse) if ID==`number' replace y2013_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2013_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2013_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2013_99_L_2=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(4/5).Percent if ID==`number', lag(4) estimates table ., star(.1 .05 .01) drop temp 57 } capture noisily reg Percent L(5/6).Percent if ID==`number',r if _rc==0 { replace AIC_2 = ln(_result(4)/_result(1))*4+(1+_result(3))*2 if ID==`number' & Year==2014 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2014_2=temp if ID==`number' replace y2014U_2=temp+1.96*e(rmse) if ID==`number' replace y2014L_2=temp-1.96*e(rmse) if ID==`number' replace y2014_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2014_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2014_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2014_99_L_2=temp-2.576*e(rmse) if ID==`number' capture noisily newey Percent L(5/6).Percent if ID==`number', lag(5) estimates table ., star(.1 .05 .01) estat durbinalt drop temp } capture noisily reg Percent L(6/7).Percent if ID==`number',r if _rc==0 { replace AIC_2 = ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2015 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2015_2=temp if ID==`number' replace y2015U_2=temp+1.96*e(rmse) if ID==`number' replace y2015L_2=temp-1.96*e(rmse) if ID==`number' replace y2015_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2015_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2015_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2015_99_L_2=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(6/7).Percent if ID==`number', lag(6) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(7/8).Percent if ID==`number',r if _rc==0 { replace AIC_2 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2016 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2016_2=temp if ID==`number' replace y2016U_2=temp+1.96*e(rmse) if ID==`number' replace y2016L_2=temp-1.96*e(rmse) if ID==`number' replace y2016_90_U_2=temp+1.645*e(rmse) if ID==`number' replace y2016_90_L_2=temp-1.645*e(rmse) if ID==`number' replace y2016_99_U_2=temp+2.576*e(rmse) if ID==`number' replace y2016_99_L_2=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(7/8).Percent if ID==`number', lag(7) estimates table ., star(.1 .05 .01) drop temp 58 } *** Three Period Lag *** capture noisily reg Percent L(1/3).Percent if ID==`number',r if _rc==0 { replace AIC_3=ln(_result(4)/_result(1))*7+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2010_3=temp if ID==`number' replace y2010U_3=temp+1.96*e(rmse) if ID==`number' replace y2010L_3=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U_3=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L_3=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U_3=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L_3=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp } capture noisily reg Percent L(2/4).Percent if ID==`number',r if _rc==0 { replace AIC_3 = ln(_result(4)/_result(1))*6+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011_3=temp if ID==`number' replace y2011U_3=temp+1.96*e(rmse) if ID==`number' replace y2011L_3=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U_3=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L_3=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U_3=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L_3=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(2/4).Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(3/5).Percent if ID==`number',r if _rc==0 { replace AIC_3 = ln(_result(4)/_result(1))*5+(1+_result(3))*2 if ID==`number' & Year==2012 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2012_3=temp if ID==`number' replace y2012U_3=temp+1.96*e(rmse) if ID==`number' replace y2012L_3=temp-1.96*e(rmse) if ID==`number' replace y2012_90_U_3=temp+1.645*e(rmse) if ID==`number' replace y2012_90_L_3=temp-1.645*e(rmse) if ID==`number' replace y2012_99_U_3=temp+2.576*e(rmse) if ID==`number' replace y2012_99_L_3=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(3/5).Percent if ID==`number', lag(3) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(4/6).Percent if ID==`number',r 59 if _rc==0 { replace AIC_3 = ln(_result(4)/_result(1))*4+(1+_result(3))*2 if ID==`number' & Year==2013 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2013_3=temp if ID==`number' replace y2013U_3=temp+1.96*e(rmse) if ID==`number' replace y2013L_3=temp-1.96*e(rmse) if ID==`number' replace y2013_90_U_3=temp+1.645*e(rmse) if ID==`number' replace y2013_90_L_3=temp-1.645*e(rmse) if ID==`number' replace y2013_99_U_3=temp+2.576*e(rmse) if ID==`number' replace y2013_99_L_3=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(4/6).Percent if ID==`number', lag(4) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(5/7).Percent if ID==`number',r if _rc==0 { replace AIC_3 = ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2014 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2014_3=temp if ID==`number' replace y2014U_3=temp+1.96*e(rmse) if ID==`number' replace y2014L_3=temp-1.96*e(rmse) if ID==`number' replace y2014_90_U_3=temp+1.645*e(rmse) if ID==`number' replace y2014_90_L_3=temp-1.645*e(rmse) if ID==`number' replace y2014_99_U_3=temp+2.576*e(rmse) if ID==`number' replace y2014_99_L_3=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(5/7).Percent if ID==`number', lag(5) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(6/8).Percent if ID==`number',r if _rc==0 { replace AIC_3 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2015 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2015_3=temp if ID==`number' replace y2015U_3=temp+1.96*e(rmse) if ID==`number' replace y2015L_3=temp-1.96*e(rmse) if ID==`number' replace y2015_90_U_3=temp+1.645*e(rmse) if ID==`number' replace y2015_90_L_3=temp-1.645*e(rmse) if ID==`number' replace y2015_99_U_3=temp+2.576*e(rmse) if ID==`number' replace y2015_99_L_3=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(6/8).Percent if ID==`number', lag(6) estimates table ., star(.1 .05 .01) drop temp } *** Four Period Lag *** 60 capture noisily reg Percent L(1/4).Percent if ID==`number',r if _rc==0 { replace AIC_4=ln(_result(4)/_result(1))*6+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2010_4=temp if ID==`number' replace y2010U_4=temp+1.96*e(rmse) if ID==`number' replace y2010L_4=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U_4=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L_4=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U_4=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L_4=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp } capture noisily reg Percent L(2/5).Percent if ID==`number',r if _rc==0 { replace AIC_4 = ln(_result(4)/_result(1))*5+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011_4=temp if ID==`number' replace y2011U_4=temp+1.96*e(rmse) if ID==`number' replace y2011L_4=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U_4=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L_4=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U_4=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L_4=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(2/5).Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(3/6).Percent if ID==`number',r if _rc==0 { replace AIC_4 = ln(_result(4)/_result(1))*4+(1+_result(3))*2 if ID==`number' & Year==2012 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2012_4=temp if ID==`number' replace y2012U_4=temp+1.96*e(rmse) if ID==`number' replace y2012L_4=temp-1.96*e(rmse) if ID==`number' replace y2012_90_U_4=temp+1.645*e(rmse) if ID==`number' replace y2012_90_L_4=temp-1.645*e(rmse) if ID==`number' replace y2012_99_U_4=temp+2.576*e(rmse) if ID==`number' replace y2012_99_L_4=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(3/6).Percent if ID==`number', lag(3) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(4/7).Percent if ID==`number',r if _rc==0 { 61 replace AIC_4 = ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2013 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2013_4=temp if ID==`number' replace y2013U_4=temp+1.96*e(rmse) if ID==`number' replace y2013L_4=temp-1.96*e(rmse) if ID==`number' replace y2013_90_U_4=temp+1.645*e(rmse) if ID==`number' replace y2013_90_L_4=temp-1.645*e(rmse) if ID==`number' replace y2013_99_U_4=temp+2.576*e(rmse) if ID==`number' replace y2013_99_L_4=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(4/7).Percent if ID==`number', lag(4) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(5/8).Percent if ID==`number',r if _rc==0 { replace AIC_4 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2014 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2014_4=temp if ID==`number' replace y2014U_4=temp+1.96*e(rmse) if ID==`number' replace y2014L_4=temp-1.96*e(rmse) if ID==`number' replace y2014_90_U_4=temp+1.645*e(rmse) if ID==`number' replace y2014_90_L_4=temp-1.645*e(rmse) if ID==`number' replace y2014_99_U_4=temp+2.576*e(rmse) if ID==`number' replace y2014_99_L_4=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(5/8).Percent if ID==`number', lag(5) estimates table ., star(.1 .05 .01) drop temp } *** Five Period Lag *** capture noisily reg Percent L(1/5).Percent if ID==`number',r if _rc==0 { replace AIC_5=ln(_result(4)/_result(1))*5+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' estimates table ., star(.1 .05 .01) replace y2010_5=temp if ID==`number' replace y2010U_5=temp+1.96*e(rmse) if ID==`number' replace y2010L_5=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U_5=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L_5=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U_5=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L_5=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp } capture noisily reg Percent L(2/6).Percent if ID==`number',r if _rc==0 { 62 replace AIC_5 = ln(_result(4)/_result(1))*4+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011_5=temp if ID==`number' replace y2011U_5=temp+1.96*e(rmse) if ID==`number' replace y2011L_5=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U_5=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L_5=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U_5=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L_5=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(2/6).Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(3/7).Percent if ID==`number',r if _rc==0 { replace AIC_5 = ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2012 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2012_5=temp if ID==`number' replace y2012U_5=temp+1.96*e(rmse) if ID==`number' replace y2012L_5=temp-1.96*e(rmse) if ID==`number' replace y2012_90_U_5=temp+1.645*e(rmse) if ID==`number' replace y2012_90_L_5=temp-1.645*e(rmse) if ID==`number' replace y2012_99_U_5=temp+2.576*e(rmse) if ID==`number' replace y2012_99_L_5=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(3/7).Percent if ID==`number', lag(3) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(4/8).Percent if ID==`number',r if _rc==0 { replace AIC_5 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2013 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2013_5=temp if ID==`number' replace y2013U_5=temp+1.96*e(rmse) if ID==`number' replace y2013L_5=temp-1.96*e(rmse) if ID==`number' replace y2013_90_U_5=temp+1.645*e(rmse) if ID==`number' replace y2013_90_L_5=temp+1.645*e(rmse) if ID==`number' replace y2013_99_U_5=temp+2.576*e(rmse) if ID==`number' replace y2013_99_L_5=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(4/8).Percent if ID==`number', lag(4) estimates table ., star(.1 .05 .01) drop temp } *** Six Period Lag *** capture noisily reg Percent L(1/6).Percent if ID==`number',r 63 if _rc==0 { replace AIC_6 = ln(_result(4)/_result(1))*4+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2010_6=temp if ID==`number' replace y2010U_6=temp+1.96*e(rmse) if ID==`number' replace y2010L_6=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U_6=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L_6=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U_6=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L_6=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp } capture noisily reg Percent L(2/7).Percent if ID==`number',r if _rc==0 { replace AIC_6 = ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011_6=temp if ID==`number' replace y2011U_6=temp+1.96*e(rmse) if ID==`number' replace y2011L_6=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U_6=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L_6=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U_6=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L_6=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(2/7).Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } capture noisily reg Percent L(3/8).Percent if ID==`number',r if _rc==0 { replace AIC_6 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2012 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2012_6=temp if ID==`number' replace y2012U_6=temp+1.96*e(rmse) if ID==`number' replace y2012L_6=temp-1.96*e(rmse) if ID==`number' replace y2012_90_U_6=temp+1.645*e(rmse) if ID==`number' replace y2012_90_L_6=temp-1.645*e(rmse) if ID==`number' replace y2012_99_U_6=temp+2.576*e(rmse) if ID==`number' replace y2012_99_L_6=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(3/8).Percent if ID==`number', lag(3) estimates table ., star(.1 .05 .01) drop temp } *** Seven Period Lag *** capture noisily reg Percent L(1/7).Percent if ID==`number',r if _rc==0 { 64 replace AIC_7=ln(_result(4)/_result(1))*3+(1+_result(3))*2 if ID==`number' & Year==2010 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2010_7=temp if ID==`number' replace y2010U_7=temp+1.96*e(rmse) if ID==`number' replace y2010L_7=temp-1.96*e(rmse) if ID==`number' replace y2010_90_U_7=temp+1.645*e(rmse) if ID==`number' replace y2010_90_L_7=temp-1.645*e(rmse) if ID==`number' replace y2010_99_U_7=temp+2.576*e(rmse) if ID==`number' replace y2010_99_L_7=temp-2.576*e(rmse) if ID==`number' estat durbinalt drop temp } capture noisily reg Percent L(2/8).Percent if ID==`number',r if _rc==0 { replace AIC_7 = ln(_result(4)/_result(1))*2+(1+_result(3))*2 if ID==`number' & Year==2011 estimates table ., star(.1 .05 .01) predict temp if ID==`number' replace y2011_7=temp if ID==`number' replace y2011U_7=temp+1.96*e(rmse) if ID==`number' replace y2011L_7=temp-1.96*e(rmse) if ID==`number' replace y2011_90_U_7=temp+1.645*e(rmse) if ID==`number' replace y2011_90_L_7=temp-1.645*e(rmse) if ID==`number' replace y2011_99_U_7=temp+2.576*e(rmse) if ID==`number' replace y2011_99_L_7=temp-2.576*e(rmse) if ID==`number' estat durbinalt capture noisily newey Percent L(2/8).Percent if ID==`number', lag(2) estimates table ., star(.1 .05 .01) drop temp } *** Eight Period Lag, but just to show it gives no useful data *** capture noisily reg Percent L(1/8).Percent if ID==`number',r } by ID : ipolate Percent_1 Year, generate(yr) epolate 65