250 likes | 406 Views
SIS 2009 – Statistical Conference. Statistical Methods for the Analysis of Large Data-sets . Generational Determinants on the Employment Choice in Italy. Chieti - Pescara, Italy – September 23 - 25, 2009. Claudio QUINTANO Rosalia CASTELLANO Gennaro PUNZO. AIM OF THE WORK.
E N D
SIS 2009 – Statistical Conference Statistical Methods for the Analysis of Large Data-sets Generational Determinants on the Employment Choice in Italy Chieti - Pescara, Italy – September 23 - 25, 2009 Claudio QUINTANO Rosalia CASTELLANO Gennaro PUNZO University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
AIM OF THE WORK Exploring some crucial factors playing a significant role in employment decision-making in Italy (GENERATIONAL PERSPECTIVE) • INFLUENCE OF “FAMILY BACKGROUND” ON THE CHOICE TO BECOME A “SELF-EMPLOYED” RATHER THAN A “SALARIED” • IF THE OCCUPATIONAL CHOICE IS “CONTEXT-DEPENDENT” (ENVIRONMENTAL ATTRIBUTES ARE ALSO MODELLED) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
SELF-EMPLOYED ARE INDIVIDUALS WHO EARN NO WAGE OR SALARY BUT WHO DERIVE INCOME BY EXERCISING THEIR PROFESSION OR BUSINESS ON THEIR OWN ACCOUNT AND AT THEIR OWN RISK (PARKER, 2004) SINCE 1990s, THE RATIO BETWEEN SELF-EMPLOYED WORKERS AND WAGE EARNERS HAS SUBSTANTIALLY BEEN STEADY OVER TIME IN THE LAST YEARS, SELF-EMPLOYMENT HAS BEEN GROWING IN SEVERAL DEVELOPED OR DEVELOPING ECONOMICS, ALSO DUE TO FLOURISHING OF SOME INNOVATIVE NON-STANDARD KINDS OF WORK SINCE THE END OF 1970s, ITALY HAS KEPT A SIGNIFICANT INCIDENCE OF SELF-EMPLOMENT (ROUGHLY 20-25%), CONSISTENTLY HIGHER THAN THE UE AVERAGE PERSONAL ATTITUDES OR REQUIREMENTS FAMILY-SPECIFIC BACKGROUND SOCIO-ECONOMIC ENVIRONMENT University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
DATA SOURCES – SHIW • Cross-sectional data come from the biennial Survey on Household’s Income and Wealth (wave 2006) • SHIW has been carrying out in Italy since 1965 by Bank of Italy • A two-stage sampling design with a partial overlap (since 1987) MUNICIPALITIES STRATIFIED by demographic size, inside each NUTS2 region HOUSEHOLDS WITHIN EACH STRATUM drawn from municipality-registers by a Simple Random Sampling Self-Representing Municipalities > 40.000 inhabitants (7.000 – 8.000 household units for each wave) Non Self-Representing Municipalities < 40.000 inhabitants – PPS University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
DATA SOURCES – SHIW • With regard to wave 2006, the SAMPLE SIZE consists of: • 7.768 HOUSEHOLDS and • 19.551 INDIVIDUALS • It is a valuable data source on a substantial range of socio-economic topics both at “household” and “individual” level • Particularly, it provides detailed information on Employment Status and Sector of Activity as well as on the single Income and Wealth components over the whole previous calendar year • Most importantly, it also detects a set of “RETROSPECTIVE PARENTAL INFORMATION”, such as: • Educational qualification • Employment status • Sector of activity allowing to account for potential generational changes over time University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLANATORY ANALYSIS WAGE-EMPLOYED WORKERS 30.95% 61.39% 7.66% NOT-EMPLOYED INDIVIDUALS SELF-EMPLOYED WORKERS University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLORING SELF-EMPLOYMENT IN ITALY • In this work, we focus on all those individuals (7.510) who are either (irrespective of their activity sector): 19.88% • SELF-EMPLOYED workers (1.493) or • WAGE-EMPLOYED workers (6.017) 80.12% Self-Employment = 0.2481 Wage-Employment 31.21% • Self-Employed by SECOND (466) or • by FIRST GENERATION (1.027) 68.79% SE by Second Generation = 0.4537 SE by First Generation University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLORING SELF-EMPLOYMENT IN ITALY Self-Employed workers by Occupational Status and “Generational Stage” 41.4% Source: Authors’ elaborations on SHIW data (2006) • The Relative Distribution of self-employed workers by FIRST Generation basically reproduces that one of self-employed workers’, taken as a whole • More than 40% of self-employed workers by SECOND Generation had (or has) at least a parent occupied in the same occupation (essentially, own-account worker/craft workers) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLORING SELF-EMPLOYMENT IN ITALY Main individual characteristics by Employment Status: summary statistics (1) Standard deviations in parentheses Source: Authors’ elaborations on SHIW data (2006) • A higher concentration of self-employment amongst HHs, while HH’s spouses/partners seem to be more “wage-oriented” • A higher incidence of men in self-employment than their wage-and-salary counterparts University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLORING SELF-EMPLOYMENT IN ITALY Main individual characteristics by Employment Status: summary statistics (2) Standard deviations in parentheses Source: Authors’ elaborations on SHIW data (2006) • On average, self-employed workers tend to be “older” than their wage-and-salary counterparts: - The incidence of self-employment is poorer than wage-employment in the lower age-classes - It tends to be more concentrated amongst individuals in mid-career (i.e., between 35 and 44 years of age) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLORING SELF-EMPLOYMENT IN ITALY Main individual characteristics by Employment Status: summary statistics (3) Standard deviations in parentheses Source: Authors’ elaborations on SHIW data (2006) • A higher incidence of self-employed workers with a “low education” than their wage-and-salary counterparts • Income levels are surely higher for self-employed workers as well as the incidence of self-employed workers owning their main home University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
A METHODOLOGICAL VIEW In the sphere of Binary Response Models (BRM), we choose… Maximum Likelihood Logit Models (Long, 1997)... with a LATENT variable that generates the MANIFEST y’s UNOBSERVABLE CONTINUOUS Underlying decisional process, essentially based on comparison between the utilities of the two employment status, leads out to the choice to become a self-employed (S) rather than a salaried (E) OUTCOME BINARY • 1 if self-employed (S) • 0 if salaried (E) “Individuals become and stay self-employed when the relative advantages are higher than in dependent employment” (Arum and Muller, 2004) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
A METHODOLOGICAL VIEW • Each individual (i ) is assumed to choose SELF-EMPLOYMENT (S) if: > • The LATENT variable (y*), that is the relative advantage to self-employment, is supposed to be linearly related to the observed x’s through the structural model (Long, 1997): • In other words, the positive outcome to become a self-employed only occurs when ; that is: or if the intercept term is not modeled University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
A METHODOLOGICAL VIEW • In this light, the probability(πi) of a positive outcome is: with i = 1, 2, ... , n • The error terms (εi) are hypothesized to obey a Standard Logistic Distribution (Long, 1997): - It is symmetric with mean zero and variance - It is remarkably similar in shape to the normal distribution • As a result, the probability of the event (y=1) is formulated in the Cumulative Standard Logistic Probability Density Function (cdf): and • BRMs are estimated by Maximum Likelihood (ML) method (Fisher’s Scoring Algorithm as optimization technique) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLANATORY VARIABLES AND STEPWISE PROCEDURE INDIVIDUAL – LEVEL COVARIATES • SOCIO-DEMOGRAPHIC VARIABLES University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLANATORY VARIABLES AND STEPWISE PROCEDURE • PROXY VARIABLES FOR THE MEASUREMENT OF WORKERS’ CAPITAL University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
EXPLANATORY VARIABLES AND STEPWISE PROCEDURE 2) AREA – LEVEL COVARIATES • Models are also enriched by a set of ENVIRONMENTAL ATTRIBUTES related to each ITALIAN REGION where the worker is LOCATED • They are selected by the Istat database of Territorial Indicators (year 2006) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
MODELS ESTIMATION Logit models with latent variable on self-employment probability (τ=0) by the whole workers’ sample and by gender: coefficients and standard errors Significance levels: *** 99%; ** 95%; * 90% Source: Authors’ elaborations on SHIW data and Istat (2006) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
HYPOTHESIS TESTING AND GOODNESS OF FIT • MODELS CONVERGENCE STATUS (by Fisher’s Scoring Method, Iteratively Reweighted LS Algorithm): GCONV = 1E-8 Satisfied 2) TESTING GLOBAL NULL HYPOTHESIS: BETA = 0 Source: Authors’ elaborations on SHIW data and Istat (2006) 3) MODEL FIT STATISTICS Source: Authors’ elaborations on SHIW data and Istat (2006) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
INTERPRETATION PROBLEMS Generally, the NON-LINEARITY of BRMs results in difficulty interpreting the effects of the independent variables (Long, 1997) RANGES OF PREDICTED PROBABILITIES: WHOLE SAMPLE [0.0423; 0.9813] [0.0461; 0.9712] - MALE GENDER SEPARATION [0.0750; 0.7684] - FEMALE The relationship between the x’s and the probability may not be approximately linear Association of Predicted Probabilities and Observed Responses Source: Authors’ elaborations on SHIW data and Istat (2006) University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
PLOTTING PREDICTED PROBABILITIES Gender vs Age Gender vs Parents’ Education Level .22 .6 Female Male Male .20 Female .18 .4 P(y=1): Predicted Probabilities of SE P(y=1): Predicted Probabilities of SE .2 .10 .1 .08 26 16 36 46 56 66 0 5 8 11 13 18 21 AGE PARENTS’ EDUCATION LEVEL • AGE and PARENTS’ EDUCATION LEVEL (if it is considered alone) have a positive effect on the probability of self-employment • The gap between the two curves shows that MALE workers are more likely to be in self-employment than FEMALES and these differences: - slightly increases as workers’ Age increases - are basically steady over all the Parents’ Education Levels University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
PLOTTING PREDICTED PROBABILITIES Citizenship vs Age Citizenship vs Parents’ Education Level .25 .6 Not Italian Italian Not Italian Italian .4 .20 P(y=1): Predicted Probabilities of SE P(y=1): Predicted Probabilities of SE .2 .0 .13 26 16 46 36 56 66 0 5 8 11 13 18 21 AGE PARENTS’ EDUCATION LEVEL Marital Status vs Age Marital Status vs Parents’ Education Level .24 .6 Not Married Married Not Married Married .4 .20 P(y=1): Predicted Probabilities of SE .18 P(y=1): Predicted Probabilities of SE .2 .15 .1 .14 26 16 36 46 56 66 0 5 8 11 13 18 21 AGE PARENTS’ EDUCATION LEVEL University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
MARGINAL AND DISCRETE CHANGES • In BRMs, the SLOPE of the probability curve is not constant • It depends on the values of the independent variable of interest and other independent variables MARGINAL CHANGES For one unit increase in xk from the baseline, the probability of an event is expected to increase/decrease by the magnitude of MARGINAL CHANGE, holding all other variables constant (usually at mean values) DISCRETE CHANGES For a change in xk from xk to xk+δ, the probability of an event is expected to change by the magnitude of DISCRETE CHANGE, holding all other variables at the given values University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
MARGINAL AND DISCRETE CHANGES Marginal and Discrete Changes for some explanatory variables Source: Authors’ elaborations on SHIW data and Istat (2006) • For instance, for 1 increase in workers’ age, the probability to be in self-employment is expected to increase by .0014 • For a dummy variable – i.e., the gender – if a worker is male, his probability to be in self-employment is .0744 greater than a female worker, holding all other variables at the given values University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research
CONCLUDING REMARKS AND FURTHER DEVELOPMENTS The direct evidence points clearly to strong intergenerational links between “parents” and “”: Parental self-employment undoubtedly increases the probability to be “OCCUPATIONAL FOLLOWERS” Having both self-employed parents strongly affects the chance to enter (or to transit) in self-employment Looking at the extent of intergenerational im-mobility from the occupational point of view: Potential differentials of “Employment Patterns” across Italian regions through small area models (territorial dimension) Extensive international comparison in analysing the different profiles in employment status (cross-country perspective) Investigating in-depth the quite heterogeneous self-employment works in order to test significantly differences within and amongst them University of Naples “Parthenope” – Faculty of Economics Department of Statistics and Mathematics for Economic Research