The fourth Portuguese Stata Users Group meeting was Friday, 15 September 2017 at the University of Porto, but you can view the program and presentation slides below.
Using Stata to estimate nonlinear models with high-dimensional fixed effects
Abstract: The community-contributed command reghdfe provides the Stata community with an excellent tool for estimation of linear regression models with multiple fixed effects. In this talk, I show that by writing simple wrappers around reghdfe, one can estimate some nonlinear models with multiple fixed effects. Basically, this approach can be applied to all models that can be estimated recursively by least squares, such as those that can be estimated by iteratively reweight least squares (Poisson, logit, probit, etc.) or by nonlinear least squares.
University of Porto and Banco de Portugal
Educational mismatches and wages: Evidence from a matched employer–employee dataset
Abstract: Skills mismatch in the labor market is a matter of great concern for academics, practitioners, and policymakers. Skills mismatch arise from several imbalances between the skills offered and the skills demanded in the labor market because of information asymmetries, transaction costs, and unresponsive education and training systems to the world of work (Quintini 2011; ILO 2014). These imbalances, whether temporary or permanent, lead to inefficiencies in the utilization of labor, with detrimental effects on productivity and growth. Interest on this issue emerged in the 1970’s because of a boom in the supply of graduates in the U.S., with Freeman (1976) as one of the first economists to express concern on the potential problem of overinvestment on college education. This discussion became, in recent years, also a topic of concern in European countries, reinforced by the huge massification of higher education enhanced by the Bologna process that favored the increase in educational attainment. In this line of research, there is a growing body of literature that aims to study the effects of overeducation on wages and other labor market outcomes (see Leuven and Oosterbeek, 2011 for an insightful literature review).
An ILO (2014) report showed that the level of skills mismatch is considerable in Europe (between 30 and 50% of the employed in the European countries are mismatched), with overeducation rising and undereducation decreasing in the majority of the countries studied.
In this paper, we will examine educational mismatches among the employed and their effects on wages. Exploring a rich matched employer–employee dataset and focusing on vertical mismatches in the Portuguese labor market over the period 1995–2012, the main goal of this study is to investigate to what extent job-mismatched workers suffer a penalty on wages when compared with similar well job-matched workers. Our data come from Quadros de Pessoal (hereafter, QP), a large longitudinal-linked employer–employee administrative dataset collected by the Portuguese Ministry of Employment. QP covers virtually all firms operating in the Portuguese private sector and employing at least one wage earner. Available information at the firm level includes employment, sales, industry, ownership, and locations. At the individual level, QP reports information on each worker’s age, education, gender, qualifications, wages, occupation, tenure, number of hours worked, and type of contract. All firms, establishments, and workers are identified with a unique identification number, so they can be matched and followed over time.
The criteria used to define educational mismatch among the employed is crucial to our analysis, and previous literature showed that the patterns of skills mismatch strongly depend on the criteria adopted to measure mismatches (ILO 2014). To identify vertical mismatches, we will rely on statistical measures based on realized matches. A vertical mismatch occurs when the level of education/qualification is higher or lower than the one required for the job. This definition is the most commonly used in the literature that studies the impact of overeducation on wages (Duncan and Hoffman 1981; Verdugo and Verdugo, 1989; Oliveira, Santos, and Kiker 2000; Hartog and Groeneveld 2004; Korpi and Tahlin 2009). Following previous studies, required education is defined as the mean or mode level education for a three-digit occupation. Then, required education for a given occupation is compared with the actual level of schooling attained by the worker in that same occupation in order to classify the individuals as over or under educated (e. g., Kiker, Santos, and Oliveira 1997). Based on these indicators of the individual’s educational mismatch status, we will estimate a mincerian wage equation that controls for workers observed and unobserved heterogeneity and firm and job title observed characteristics.
Our results show that more than 50% of the Portuguese workers in the private sector suffer from an educational job mismatch. Furthermore, regardless of the criteria used to measure over- and undereducation, the OLS results show that, on one hand, when compared with their coworkers with similar characteristics who are adequately educated, overeducated workers receive a wage bonus for the extra years of schooling, and undereducated workers a wage penalty for the extra years of deficit education. On the other hand, overeducated workers earn less and undereducated workers earn more than similar workers with the same years of schooling but who hold jobs for which they are adequately educated. These results are, in general, in accordance with previous related literature. However, the fixed-effects results indicate that taking into account workers unobserved (permanent) heterogeneity reduces considerably the discrepancy between the wages of well job-matched workers and job-mismatched workers, evidence that failure to control for individual unobserved heterogeneity may overestimate the impact of over- and undereducation on earnings.
University of Porto and cef.up
University of Porto and cef.up
Using panelstat to compute statistics for panel data
Abstract: Panel data has been widely used in economics. In this presentation we will present panelstat, a community-contributed Stata command that analyzes a panel dataset and produces a full characterization of the panel structure. This command allows one to check the most common patterns of the data, characterize the temporal gap structure of the dataset and easily compute statistics along the panelvar or the timevar dimensions. panelstat is also a useful tool to signal abnormal absolute and relative changes over time or movements of individuals across units of a certain variable. In this presentation, we will provide some practical examples using panelstat.
Banco de Portugal
Returns to postgraduate education in Portugal: Holding on to a higher ground?
University of Minho
Hugo Figueiredo, João Cerejeira, Miguel Portela, Carla Sá, Pedro Teixeira
University of Minho
The health production function revisited: The role of social networks and liquid wealth
Abstract: Building upon the Grossman model (1972), we propose an extended model of health production, which accounts for the role of social network interactions and share of liquid wealth. The model predicts that both factors have a positive impact on health production. A recursive system that controls for potential sources of endogeneity of social network contacts, share of liquid wealth, and healthcare demand is used to empirically test the theoretical predictions. The estimation results show that the share of liquid wealth directly affects health in a positive and statistically significant way. Social networks do not have a direct impact on health production, though the model suggests that they indirectly enhance health through a greater use of ne cessary healthcare services. Lastly, the empirical model shows that social networks and the share of liquid wealth act as substitutes in the production of health.
Nova School of Business and Economics
Introduction to Bayesian Analysis in Stata
Abstract: Bayesian regression analysis is a growing topic of interest for researchers in different areas because of the variety of models that can be accommodated within this theoretical framework. I will outline the main aspects associated with Bayesian regression in Stata, and I will show the new facilities incorporated in Stata 15 to make this kind of analysis more accessible to users in different disciplines.
School and teacher characteristics versus student progress
University of Minho
Carla Sá, Miguel Portela
University of Minho
New features in Stata 15
Abstract: A brief overview of the new features of Stata 15. I will be discussing the newest features in what StataCorp President, Bill Gould, calls "our most remarkable release yet."
Wishes and grumbles
The logistics organizer for the 2017 Portuguese Stata Users Group meeting is
the distributor of Stata in Portugal.
View the proceedings of previous Stata Users Group meetings.