Publicado
2015-12-30

Proyección de la población universitaria utilizando el modelo Lee-Carter

Forecasting college populations via Lee-Carter models

DOI: https://doi.org/10.15332/s2027-3355.2015.0002.03
Norman Diego Giraldo Gómez
Carlos Ochoa Molina

Resumen (es)

Los problemas de la deserción estudiantil y de la permanencia prolongada en el plan de estudios son de principal interés para las directivas universitarias. Una correcta medición e interpretación son necesarias para monitoriar los efectos de diferentes políticas, cambios en los reglamentos estudiantiles, etc. implementadas con el fin de resolver ambos problemas. Para lograr esto una alternativa puede consistir en establecer un modelo estadístico para la evolución de las poblaciones universitarias a través de diferentes cohortes, que es el objetivo de este trabajo. El modelo aplicado se conoce como el modelo Lee-Carter, y puede localizarse dentro del conjunto de modelos de tablas de vida dinámicas'', en el área de análisis de supervivencia actuarial. En este trabajo se aplicaron dos metodologías de ajuste para el modelo.  Se compararon los estimadores del modelo Lee Carter por el método original de descomposición singular y por el método de maxima verosimilitud en un modelo log-bilineal Poisson, utilizando la librería gnm de R. Aunque están disponibles otras librería como LifeMetrics e ilc, el procedimiento en gnm, para modelos no lineales generalizados parece ser el más aceptado en la actualidad. En particular, se implementó una regresión no-paramétrica Loess para el parámetros $\kappa_t$ para pronosticarlo a corto plazo, y se comparó con un modelo  ARIMA(1,1,0) con tendencia (los modelos ARIMA se recomiendan en la literatura para la estimación de $\kappa_t$). En ambos  casos se observó una evolución estocástica del parámetro $\kappa_t$ durante el período entre los semestres 1989-02 a 2006-01, en total 34 semestres. Con los modelos ARIMA(1,1,0) y Loess se procedió a calcular 10 pronósticos para $\kappa_t$. Con el pronóstico número 10, correspondiente a la cohorte del semestre 2011-01, se calculó la evolución futura de esta cohorte durante 17 semestres, de los cuales ya se han observado 8 al final del año 2014. Al comparar los observados con los calculados los resultados son satisfactorios y los pronósticos coinciden razonablemente con los valores observados, por lo que los restantes pronósticos son confiables. La evolución de la mortalidad durante 34 semestres (17 años) permite obtener conclusiones acerca de los cambios en la deserción y egreso de la población estudiantil, las cuales aparecen en la última sección del estudio.

Palabras clave (es): modelo Lee-Carter, poblaciones universitarias, proyección de poblaciones, tablas de vida.

Resumen (en)

In Colleges with large student populations, it is important to have a system for monitoring their evolution along the time. Proper measurement and interpretation are necessary to help evaluate the effects of different policies, changes in student regulations, medical costs, insurance, student welfare, etc. To achieve these goals an alternative may be to establish a statistical model for the population evolution across different cohorts, which is the subject of this article. The model used is known as the Lee-Carter model, which can be seen as an example of “dynamic life tables models”, in the area of actuarial survival analysis. In this work we applied two estimationg methodologies for adjustment of this model to real life data, from a Colombian University. The estimators were the original method of singular decomposition and the method of maximum likelihood in a log-bilinear Poisson model, using the library of gnm in R. Although there are other available libraries like LifeMetrics and ilc, the procedure in the gnm library, for generalized non-linear models, seems to be the most accepted at present. Additionally, we implemented a non-parametric regression of Loess type for one of the parameters in the LeeCarter model: the κt, which enabled us to calculate forecasts for predicting the central rate of mortality; we compared this forecast with the ones obtained with a model ARIMA(1,1,0) with trend, given that the ARIMA models are recommended in the literature for the forecasting of κt. Models with the ARIMA(1,1,0 ) and loess is proceeded to calculate 10 forecasts for κt. With the forecast number 10, corresponding to the cohort semester 2011-01, we estimated the future evolution of this cohort during 17 semesters, of which had already been observed the first 8 at the end of the year 2014. When comparing the observed populations with the calculated ones results are satisfactory, and the forecasts match reasonably well with the observed values, so that the remaining forecasts are reliable. The evolution of mortality during 34 semesters (17 years) lets us obtain conclusions about the changes in the student population, which appear in the last section of the study
Norman Diego Giraldo Gómez, Universidad Nacional de Colombia, sede Medellín

escuela de estadistica

universidad nacional de colombia

matematico (UN), ms en matematicas (UN)

profesor asociado

Referencias

Booth, H., Hyndman, R. & Tickle, L. (2013), `Prospective life tables', url:robjhyndman.com/papers/prospect.pdf.

Bowers, N., Gerber, H., Hickman, J., Jones, D. & Nesbitt, C. J. (1986), Actuarial Mathematics, The Society of Actuaries, Ithasca, Il

Brouhns, N. & Denuit, M. (2002), `Risque de longeevite et rentes viageres II.Tables de mortalit e prospectives pour la population belge',Belgian ActuarialBulletin2(1), 50-63.

Brouhns, N., Denuit, M. & Vermunt, J. (2002), `A Poisson log-bilinear regression approach to the construction of projected lifetables', Insurance: Mathematics and Economics31, 373-393.

Butt, Z. & Haberman, S. (2009), `Ilc: A Collection of R Functions for Fittinga Class of Lee-Carter Mortality Models using Iterative Fitting Algorithms', Actuarial Research Paper No. 190,Faculty of Actuarial Science and Insurance, Cass Business Schoolpp. 1-37.

Cairns, A. J. G. (2007), `LifeMetrics: A toolkit for measuring and managing longevity and mortality risk',url:www.lifemetrics.com.

Cairns, A. J. G., Blake, D., Dowd, K., Coughlan, G. D., Epstein, D., Ong, A.& Balevich, I. (2009), `A Quantitative Comparison of Stochastic Mortality Models Using Data from England and Wales and the United States', North American Actuarial Journal13(1), 1-35.

Cleveland, W. S., Grosse, E. & Shyu, W. M. (1992),Local regression models, In:J.M. Chambers and T.J. Hastie, Statistical Models in S, Wadsworth and Brooks/Cole, Chapter 8.

Currie, I. (2013), `Smoothing constrained generalized linear models with an application to the Lee-Carter model', Statistical Modeling13(1), 69-93.

Currie, I. (2014), `Ontting generalized linear and nonlinear models of mortality', Scandinavian Actuarial Journal14, 1-28.

Davidian, M. (2009), Nonlinear models for univariate and multivariate response, Lecture Notes. Department of Statistics., North Caroline State University.

Delwarde, A., Denuit, M. & Eilers, P. (2007), `Smoothing the Lee-Carter and Poisson log-bilinear models for mortality forecasting: a penalized likelihood approach', Statistical Modelling7, 29-48.

Eckart, C. & Young, G. (1936), `The approximation of one matrix by another of lower rank',Psychometrika1, 211-218.

Huertas, J. A. (2001),C~A

Hyndman, R. (2014),forecast: Forecasting functions for time series and linear models. R package version 5.5.*http://CRAN.R-project.org/package=forecast

Hyndman, R., Booth, H., Tickle, L. & Maindonald, J. (2014),demography: Fore-casting mortality, fertility, migration and population data. R package version1.17.*http://CRAN.R-project.org/package=demography

Koissi, M.-C. & Shapiro, A. (2008), `The Lee-Carter model under the condition of variables age-specific parameters',43rd Actuarial Research Conference, Regina, Canada.

Lee, R. & Carter, L. R. (1992), `Modelling and Forecasting U.S. Mortality', Journal of the American Statistical Association87, 659-671.

R Development Core Team (2008),R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria. ISBN3-900051-07-0.*http://www.R-project.org

Renshaw, A. E. & Haberman, S. (2009), `A cohort-based extension to the Lee-Carter model for mortality reduction factors', Insurance Mathematics and Economics.

Shores, T. (2004),Applied Linear Algebra and Matrix Analysis, Springer Verlag, Heildelberg.

Steinsaltz, D. (2010),Statistical Life-Time Models. Lecture Notes, Department of Statistics, Oxford University.*http://www.steinsaltz.me.uk/BS3b/BS3b.html

Turner, H. & Firth, D. (2007),Generalized nonlinear models in R: An overviewof the gnm package, ESRC National Centre for Research Methods, NCRMWorking Paper Series 6/07.*http://www.ncrm.ac.uk

Wilmoth, J. (1993), `Computational methods fort ting and extrapolating the Lee-Carter model of mortality change', Technical Report. Department of De-mography, University of California, Berkeley.

Cómo citar

Gómez, N. D. G., & Ochoa Molina, C. (2015). Proyección de la población universitaria utilizando el modelo Lee-Carter. Comunicaciones En Estadística, 8(2), 173-192. https://doi.org/10.15332/s2027-3355.2015.0002.03