Outline for ECON 753 at UMass(2)
Outline for ECON 753 at UMass(2)
Popular in Course
Popular in Department
This 45 page Class Notes was uploaded by an elite notetaker on Friday February 6, 2015. The Class Notes belongs to a course at University of Massachusetts taught by a professor in Fall. Since its upload, it has received 27 views.
Reviews for Outline for ECON 753 at UMass(2)
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 02/06/15
Applied Econometrics Labor Econometrics Michael Ash Econ 753 Labor Econometrics 7 p145 Outline of Presentation Threeday plan 1 Wage models 2 Labor demand the employment effect of the minimum wage 3 Labor supply the employment effect Today Wage models 1 Why study wage determination 0 Outcome of a labormarket process 0 Distribution of product and surplus 39 Rents both for themselves and as an indicator of market power 2 Building an econometric model Theory and functional form 3 Discrimination 4 Aggregate variables and micro units Labor Econometrics 7 p245 Data sources and de nitions 1 Census CPS NLSY PSID US 2 Luxembourg Income Study OECD GSOEP Germany Wage and wage 0 Annual earnings weekly earnings hourly wage salaries 0 Think about the data in rows observations records and columns variables fields Read and produce tables of regression output I I Labor Econometrics 7 p345 Wagesetting models 0 Competitive market models 0 Marketclearing wage 0 Human Capital HK hedonic wage model 0 Compensating differentials 0 Monopsonistic or exploitation models 0 Institutional models and discrimination models 0 Segmented labor markets 0 Interindustry wage differentials Labor Econometrics 7 p445 On econometric models of wages DoesX orZ raise wages 0 Simultaneityendogeneityselection DoesX realy raise your wages or do people with X tend to have high wage 0 Signalling critique of Human Capital 0 General equilibrium remark on Human Capital 0 Aggregate explanatory variables Labor Econometrics 7 p545 Functional form 39 About subscripts and parameters yi y yi B 0 Interpreting the semilog specification 0 The Return to Education 39 Experience or tenure Indicator or dummy variables Indicator interactions 0 Differences in the function over timespacegroup eg skillbiased technical change Labor Econometrics 7 p645 Return to Education HK phrasing Human Capital Regressions are based on a hedonic model of wage determination the wage is a function of the valuable characteristics of the worker What s the return to attending one more year of schooling Career Earnings by Years of School Years attended t 0 1 2 0 Y0 O O 1 Y0 Y1 0 2 Y0 Y1 Y2 3 Y0 Y1 Y2 5 Y0 Y1 Yz T Y0 Y1 Y2 Labor Econometrics 7 p745 Return to Education Note that more years of schooling require foregoing years of earnings In a very simple model the return to schooling will equalize the net present value N PV of the alternatives 012 or 3 years of schooling Y0 Yo Yo 1r NPVS0 Y 1r01r1 1r2 r 0 0 Y1 Y1 1 NPV 1 Y S 1r01r11r2 r 1 Equilibrium NPVS 011 Y0 NPVS 1 implies Y1 1F1 Y0 Labor Econometrics 7 p845 Returns to Education The additional income from one year of school relative to zero years of school represents a return of m The same argument implies that Y2 1 F2 39Y1 and so on Substituting recursively we find Ys 1 rsYs1 1 rs1r1Yo For the moment assume that rs rs1 r1 r a single percent return to an additional year of education a testable proposition YS 1S Y0 Labor Econometrics 7 p945 Measuring and Estimation Recall that by a Taylor series approximation 1 5 e8 and so 1 rs ers e This might also be familiar from compoundinterest formulas Ys e Yo and taking log of both sides lnYS lnYo rs which can be interpreted as a semilog wage equation that we can estimate lnYS 06 BS YSand S are measured earnings and schooling and or and B are estimated parameters B is the return to schooling expressed in percent per year and or is the intercept implied log earnings at zero years of education I Labor Econometrics 7 p 1045 Critique of Causality In observational studies in the US and elsewhere B is rarely estimated below 005 a 5 percent return to schooling and in LDC s is sometimes estimated as high as 020 a 20 percent return to schooling ls lnYS 0c BS a causal relationship ie if B is positive does schooling cause higher earnings Why or why not 39 Omitted variables that cause B to be an overestimate left and right critiques O Socioeconomic status causes both schooling and earnings 0 Ability causes both schooling and earnings 0 Signalling General equilibrium and social returns 0 Attitudinal modification We will return to this question in more detail later Labor Econometrics 7 p1 145 Experience Next we introduce experience and allow it to have a nonlinear effect on earnings Why 0 Human capital explanations rational to accumulate HK when young rapid accumulation of HK in early years on the job OJT long time to amortize the cost of training forgetting and HK deterioration 0 Neoinstitutional explanations solve PA problem with bond matching and wage gains 39 Institutional explanations customs and norms seniority in unionized and unionavoiding workplaces limited evidence on productivity bus driver safety study Labor Econometrics 7 p 1245 Tenure and experience 39 Would be helpful to resolve some of the competing hypotheses 0 Empirical problem few datasets include measures of tenure then they do the quality is often poor for three reasons recollection bias definition ofjob and employer sampling durations Potential Experience Mincerian Experience for Jacob Mincer Potential Experience Age Years of Schooling 6 Reasonable proxy unless the aim is to make subtle points about tenure versus experience Note that regressions with Potential Experience and Schooling must omit age to prevent perfect collinearity Labor Econometrics 7 p 1345 Quadratic speci cation lnYi ocSBS ExBEX EXEBEXZ 8 BlnYi aEXl39 This term depends on the level of experience Typical values are BEX 003 and BExz 00004 The negative quadratic term means that the relationship is concave The positive linear term and the size of the two terms means that the return begins positive about 3 percent per year for a new worker at 0 years of experience and then falls You can compute the return at any given level of experience by substituting You can compute the peak of the ageearnings profile by setting the derivative to zero Bin aEXl39 BEX E 0 Labor Econometrics 7 p 1445 Experience continued which implies that Expeak BEX 2BEx2 For example with 35x 003 and BExz 00004 the peak of the profile would be at about 38 years of experience Labor Econometrics 7 p 1545 Dummy variables Allows different intercepts for different categories Simplest example Each observation 139 is in category D 1 or not D O eg nonwhite female or union member categorical nominal variables with a simple level effect lnYz ocDz5XzBX21 The estimated coefficient 5 is the return premium or penalty to being in category D measured in log points read as percent Tip give D a useful name eg Female is a more useful name for an I Indicator than IS Sex Nonwhite IS a more useful name for anLindigatgrsp1645 39 LL I an Interpreting dummy variables To interpret the coefficients on dummy variables Consider a very simple wage equation lnYi 06Dj5l 8i Y is hourly wage for person 139 ocis the intercept a baseline wage Diis an indicator variable for membership in category D eg D 1 for college degree or more and D1 O for less than college degree How can we interpret 8 the coefficient on D A A In 6C X Yieoe Labor Econometrics 7 p 1 745 Interpreting dummy variables For a person without a college degree D O and expected earnings are A YeoceO8eoc1eoc So the baseline wage without a college degree is ed For a person with a college degree D 1 and expected earnings are Yi eOLOel ede What s the percent return Labor Econometrics 7 p 1 845 Interpreting dummy variables YlY 0L 6 or p 16 e e e5 1 Y ea Remark this formula for percent change always works If you want to know the percent return when you are given 8 you can calculate p 66 1 But there is a shortcut when 5 is small Consider a firstorder Taylor approximation of the percent return pas a function of the coefficient 8 As we just saw p8 e8 1 p8 g p0p 0o8 O near80 p 5 65 p0 e0 11 10 p 0 60 A A A I U 1 O O 0 Labor Econometricsipl945 I The Returns to Computer Use John DiNardo and JornSteffen Pischke The Returns to Computer Use Revisited Have Pencils Changed the Wage Structure Too Quarterly Journal of Economics 112 pages 291303 1997 Also a good example of a critical replication 0 Enormous increase in inequality in the United States 0 Between categories 0 Within categories residual inequality 39 Skillbiased technical change 0 Alternatives trade tradeSBTC institutions 0 What are the technical changes in question 0 Alan Krueger offers computers I I Labor Econometrics 7 p2045 The Returns to Computer Use THE RETURNS T E39GMPU TER USE REE HATED IE itE REFER11m PM THE EFFECT OF mmmn USE cm Etin WHEN WEI mama Lam HDUFIL i WMEE l T HIiAEII Emma 24 panammmm l EF En t U 5 LT S U 5 Gasman Ge39nnnny39 Grammy m a e 193 1933 1993 19 1955 1935 WEE 11352 C mpmter ELIEI LIES DEM 112 I115 EIL39ITJ DEBS l il 39l i l l WHEN HMS1 Tears n D mum i Elv i39 LIES EILI FE l lul EIEIJEIIJEI 11092 liv m lll Em ml ll Erpverienne 113213 a iElJ 05035 Ilium lilt ll I l 1 01301 i7 quotI39l ilb 430431 i g l llj minimEmma E11343 aIil II1 l ili E HID55 I1III4F 1m ELWE39J mm21 l E MEIElir l d39 m BLUEDEE HR 44145 EFL13913 0424 13257 1123 H335 Numbur Ur aha 13335 13379 13305 IEAE T 22353 Emma The Returns to Computersiand Pencils TABLE III 018 REGRESSIDN FOR THE EFFECT OF DIFFERENT Tums 0N PAY DEPENDENT VARIABLE LOG HOURLY WAGE Ermmm Emmvs IN YARENTHESES Independent Germany Germany Germany Germany Germany Germany Germany variable 1979 1985 86 195192 1979 1979 1955 1936 1991 1992 Occupation indicann s No No ND 501 501 742 1071 Grades and father39s N0 Na Na No Yes N0 N0 Occupationquot Tools entered separamly Cnmpu39er 0112 0157 017 1 0025 0022 0076 0083 0010 0007 0000 0011 0011 0008 0007 Calculator 00087 0 128 01129 0027 0025 0061 0 054 0007 0006 0006 0008 00008 0007 0 003 Telephone 0131 00114 0136 0060 0057 0059 0072 0000 0006 0000 0007 0007 0007 0007 Penpencil 0123 0112 0127 0055 0052 0055 0050 0 000 0000 00067 0007 0 007 0007 0007 Wnrk while sitting 0106 0101 f 0042 0041 0035 0006 0007 0000 0003 0000 Hand tool 0117 0 086 01091 0048 0045 0020 0020 2g hammer7 0007 0 006 0006 0000 0009 0000 0008 LahaxEcanamemcsip 2245 Multiple dummy variables Multiple nominal categories eg race industry occupation Create a 10 dummy variable for each category Must omit one of the classes or the dummies will be collinear with the intercept Can alternatively omit the intercept The omitted class is associated with the intercept and all differences the coefficients on the dummies are read relative to the omitted class llez39 06 D22 52 i D3i53 XjBX 8239 Labor Econometrics 7 p2345 Other uses for dummy variables Can also create dummies for a small relative to N number of ordered categories a less parametric approach eg a return for every year of schooling 81 82 816 instead of restricting all years to have a single return BS Dummies are equivalent to a withingroup approach The response slope within every group is constrained to be the same but the groups can be at different levels Many dummy variables fixed effects eg for every state for every city for every person Needs more than one observation per fixed effect Cannot nest fixed effects eg neighborhoods and cities Labor Econometrics 7 p2445 Dummy interactions 39 Dummydummy interactions Interactions among categorical variables Differences in function over time place or group For example do black women suffer as much discrimination as do black men Are union premiums larger for women or for men Were union premiums larger in 1970 or in 1980 In X Femz BFem i NonWhitejBNW i Fem X NonwhiteiBFemNW 8 Show regression output For this example BFem lt 0 BNW lt 0 but BFemNW gt Oenough to offset one of the two forms of discrimination ie black women do not receive the cumulative effect of sex and race discrimination Labor Econometrics 7 p2545 Dummydummy interactions Advantages of pooling 1 restrict other coefficients to equality 2 easy to test hypotheses Interact dummy variables for each category of first characteristic with dummy variables for each category of second characteristic Include dummies level or main effect and all dummy interactions To include interactions but not the main effect you must have a good reason or model Multiple interactions are possible Preview treatmentcontrol beforeafter model identify quasiexperimental effects y or Post8pm Treatment grplISTreatment W Post gtlt Treatment gtlt 5Effect 81 SE ectis the effect of treatment on the outcome because it expresses the difference between the treatment and control groups after treatment has Labor Econometrics 7 p2645 knnn vnrsnhlnhl IAIn nn on A39F inInvnrvInhl I ll Imwul Inrinklno I Dummy interactions 0 Dummycontinuous interactions different slopes for different folks Sf 39 ggxpil III a 5 I I Labor Econometrics 7 p2745 Aggregate explanatory variables Moulton Contextual explanations of individual outcomes neighborhood city industry May want to explain an individual outcome with an aggregate characteristic eg a worker s wage may depend on the capitallabor ratio of the industry For worker z39 in industry s 1nYS or Zisy KLSB sis ns with Labor Econometrics 7 p2845 Moulton 1 BOLS is unbiased but 2 seBO39S is underestimated Standard OLS estimation will generate an unbiased estimate of B how capitallabor ratio affects wage but the standard error of B is underestimated Intuition because KLSis aggregate all workers in industry s have the same K L and the same industryspecific error ns There are effectively S not I x S observations Ignoring this understates the estimated standard error of B Good exercise Labor Econometrics 7 p2945 Moulton variables Emu L DErlmm ms pr Annmamm 5mm VMcIMsLn 39leusn IN Rmnmsmcm EMAMDLEH Variable Dci ruiifjun Economic v39ariahlci Jr I I Esm jmaicd r111 all sum mpl cmncm growth Curr mt 5m rclmth39n cmpluymtnt disturbantc 17f Predictcd 531 dislurbmxcs g Linrmr mmbina lion of t nm asts If slau helium dinlurhnnms Imam Hiawatha mu39ujclal Helmum quotuquoturinl1lrsc 1quot II in 39 WILLM In mml Lian mm mm m Litht pl lpulj l lll 914039 I 3 11 Legal almnlnm per mm like lhiNhs WEE laH1 1 Dcalh mm from heart dim355 per HJlJIlDlII popula ticn NJle 19quot Dcuth raj mm suicide pun mum Emulation l39JlZJIJ l R Ill l 39D ll l me than pawn ul minimums per Ill13111131113111 ularinn 19W x m 1 115114 of divmms apt Innu lmenls per 000 unpubl Linn 1930 l 1 it ll Pctmuting of persons 5 y cars old nrulked in public Eltmcnlnry and secondary schools W H 3 IL 3II 39l39mal Fund 111 in wu n kllmnelen X 3quot 1 Total 39h l39ll l39 sum in aqua kilmticl crs 210 quotIl Eirearisen of highcs pain in meters l K llll39jl Pu capital state lggisiativi appropriations for arts agendas Hm x 10 41 Tom numb DI black mlccl j uI ciinls July 1 h In quot1 Daily nmspupur Eitrulmimt per milpzla 1931M ll 393 Ranihm umlal Lligim I mm Eurennui WI p 5231 321111quot quot1 173 ultra 1 L lihruugh r va J w F H I durum Irmrn annual slaw mull mam11ml III39IISHIJIILTJJ EIHL IHIL HIEIIE 39I39hE 52m hEl E Labor Econometrics 7 p3045 Moulton results T ELT 2 Em1rImmn lia39Tx39lEn mr MHJHEUATE STNIT Vamwuy m m Linn m Inmvunml WFHil r Wng Ann Flap4w Im tlm WE I 2 le iem L11MJ1L3t n Adjusted f EI tiem Lln djumecl auljmmd Variable Eanimate I EMILEHE r ituiisdi Estimate P swim I statistic Emmnmic trurianhlcrs I t 1w H641 Ham LEI1 IIIIE 1 131 15quot 4343 Ll l UH 53 x EH4 131 1132 43135 i AMI Irrel m u39mi lblusz a t n c 7 rm 52 M x Ll 101 059 x l143 z 133 1 I7 EDI1 Lit LEE I mg I 515 H434 In LL15 unit III II 3 f7 339 nun 17 EL39W 204 135 g u 411411 315 i189 er 39 la1Fq 2A3 11M In Li squot If 3 1321 1 39r t 125 J344 r m I 1 39E Il l Lilli 11 2 56 413 HFI a MEI I IT a 1179 1 I p 1 111323 r 103 ics7p3145 Aggregate explanatory variables Moulton Alternative solutions 1 Estimate with standard errors adjusted for clustering 2 Estimate with dummy variables for the aggregates yis Zisy 5s i 813 Then estimate the relationship between the s premium and characteristics of aggregate s 5s st i us Labor Econometrics 7 p3245 LaborMarket Discrimination 39 What are the right questions 0 What s wrong with reverse regression 39 Oaxaca decomposition 0 Decomposition of intergroup averagewage gap into average characteristics and price per characteristic 0 Compute means for the explanatory variables for the two groups 39 Run separate regressions for the two groups 1nYW 1nYb WBW 473 WW 4pr MW 473 For B5 W mb discrimination in returns difference in characteI Labor Econometrics 7 p3345 Notes on Oaxaca decomposition 39 Note and interpret the alternative decomposition moo mm Wow 27B WBMJFBW BW b WBW BbHOW WBW 1 0 Recall thatXincludes the constant 1 1SZXjX12 Labor Econometrics 7 p3445 Labor Demand 0 Minimumwage 0 Immigration 0 Factor demand models Labor Econometrics 7 p3545 TimeSeries Minimum Wage Studies 10 percent increase in the minimum wage causes a 1 3 percent reduction in teenage employment Critiques Kaitz index MWI Emmiwile coverage and relative wage Sensitive to specification 0 Publication bias tratios do not grow with the square root of sample size 39 E MW Relationship weakened between 1950 s1970 s and 1980 s 1990 s 39 Sex oddity Labor Econometrics 7 p3645 Minimum Wage Natural Experiment Employer Responses to the Minimum Wage Evidence from the FastFood Industry The Employment Effects of the New Jersey Minimum Wage 39 Date of New Jersey minimumwage increase 1 April 1992 0 New Jersey minimum wage before 1 April 1992 425 per hour 39 New Jersey minimum wage after 1 April 1992 505 per hour 39 Federal minimum wage before and after 1 April 1992 425 per hour 39 Pennsylvania minimum wage before and after 1 April 1992 425 per hour Labor Econometrics 7 p3745 Research Design 0 Where will the minimum wage have bite Confounding factors 0 National or regional trends 0 Policy endogeneity Control or comparison groups Labor Econometrics 7 p3845 The fast food industry Approximately onehalf of employees are 20 years of age or older 66 percent of employees are female 77 percent of employees are white 65 percent of employees have at least a highschool diploma Turnover is very high onehalf of employees have less than one year of job tenure Work is gender segregated female employees and employees with higher seniority are more likely to work in the front of the store Labor Econometrics 7 p3945 Survey Method 39 Telephone survey of almost 500 Burger King KFC Wendy s and Roy Rogers restaurants in eastern Pennsylvania and New Jersey 1 3 k l l a J ansylvamn IK5Igt 271 4 39 7V 7 F ltmyquot w quoto in mu mfm 1 7 l y 39 First wave of survey FebruaryMarch 1992 87 percent response rate 39 Second wave of survey NovemberDecember 1992 100 percent of firstwave respondents 39 Fulltime equivalent FTE employment E fulltime employees onehalf of parttime employees Labor Econometrics 7 p40 45 Analysis 39 Establish that the NJ minimum wage had bite in New Jersey and not in Pennsylvania the treatment group received treatment and the comparison group did not 39 Measure the average change in employment in NJ restaurants 0 Appropriate counterfactual average change in employment in PA restaurants Labor Econometrics 7 p4145 L 1 I 2 Wm mm m mm mushy w a a m a 2 w 1 5 5 2 o LLJ Us sus 325 515 A wmmnge an E g m 2 T w 3 5 m a n39 as us as as 505 st 55 a wash NM km Pemuylvanu mm 22 D smbuuun 0 mm me am A pgbmryrma 992 E Nnvnnbeerecmlber I992 r Labnr Ecnnnmemcs 7 p 4245 Results TAEHl 22 Ain tragt Empluymen per Restaurant Before and After Increase in New Jersey Minimum Wage R mu an MS Di eremf i AH PA N NI PA H L 1393 NJ 1 F115 Employment chnre 2103 2333 2044 2351 All Availath Obsewa un 049 135 051 144 2 FTE Empluyment AltEr All 2105 21 2166 l4 Available Observations ll16 091 052 111539 3 Change in Mean FTE 005 2l 059 236 Empluyment 05m 11 25 054 136 4 Change in MEan FTE Em L39U 223 04 25 ployment Balanced Sample 046 125 143 13 of Restaurants 5 Change in Mean FIE Em 2 218 023 251 pluyment Setting FTE at 04 125 12149 135J Temporarin Closed Restau rants to 212m Labor Econometrics 7 194345 Other Outcomes Labor Econometrics 7 p4445 Labor Supply 39 Participation and hours 0 Structural models income and substitution effects 39 Natural experiment models 0 Criticisms labor demand 0 Structural models 0 Reduced formdifference in difference Labor Econometrics 7 p4545
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'