Análisis comparativo de las fuentes estadísticas para la proyección de series temporales de migraciones regionales clasificadas por niveles educativos
Résumé
This paper analyses the statistical strength of population censuses and the Labour Force Survey (EPA), with the objective of estimating regional matrixes of migratory flows classified according to education level, by means of variations of the iterative proportional fitting procedure, based on the information from such sources.
Official migration statistics do not classify migrants according to their qualifications, which makes it necessary to use alternative sources in order to model population movements among autonomous regions, taking their education into account.
The use of these alternative sources obliges us to carry out a prior analysis of such, with the objective of selecting the optimum choice enabling the best estimation of inter-regional migratory flows, taking into account the migrant population’s different education levels.
We therefore analyse the strength of migratory information from censuses and the Labour Force Survey, using different indicators and procedures, in Section 2 of this paper. This is preceded by an introduction, while Section 1 reviews the literature on qualified migrations.
In this Section 2, we analyse a first comparison of both sources with the global data from the Residential Variation Statistics (EVR). This analysis is supplemented by the results of compensatory equations for Spain as a whole and for each of its autonomous regions, based on information from the last two population censuses, calculating thereafter the mean relative errors (ERM) for the population totals and for people with second- and third-cycle studies, in each of these territories.
The conclusion of these analyses is that the mobility variable of population censuses is affected by the lack of answers from the interviewed population and by the imputation method used by the National Institute of Statistics (INE) to correct it, which explains the underestimation of census data compared to the Residential Variation Statistics.
Starting from evaluation reports on the quality of the 2001 Population Census and of the Labour Force Survey for the years 2006 to 2017, we extracted the results of several indicators such as the percentage of identically classified, the net difference rate, the net change index or the gross difference rate, concluding that in all cases there is greater strength for the Labour Force Survey data than for that of the 2001 Census.
Finally, using the Residential Variation Statistics of 2001, which includes interior migratory data classified according to academic qualifications, we calculated equality and relation indicators such as Theil’s U index of inequality, Pearson’s coefficient of variation and that of correlation, between this source and the census and this same source and the Labour Force Survey of that year. As in the case of the prior comparisons, the analysis of the previous statistics produces better results for the Labour Force Survey data compared to that of the Residential Variation Statistics, than the census data with regard to this source.
After selecting the Labour Force Survey as the statistical source for projecting regional migratory flows, taking into account education levels, in Section 3 of this paper we use the iterative proportional fitting to obtain migratory matrixes.
The procedure for estimating migratory flows based on the information from the of the Labour Force Survey micro-data is verified in several phases. Beginning with the matrix generated annually from the four files of three-monthly micro-data, filtering the population ages 16 and over that changed residence the previous year and adding the information in a matrix with 20×5 rows and 19 columns, which shows the structure of inter-regional migratory flows and of immigrants from abroad classified in five education levels. The total margins of this matrix of micro-data are used to calculate new ones by elevation, converging with the values added from the Labour Force Survey module “Subsample variables: persons that have changed residence a year ago,” using a bi-proportional adjustment procedure. In this way, the migrations classified according to educational level of NUTS-2 provided by the micro-data sample are elevated to the values of the migration movements of NUTS-1 supplied by the “Subsample variables.”
Before this estimation, the correlation of the migratory flows added from the Labour Force Survey with those of the Residential Variation Statistics are tested for all the years analysed. To that end, the data of the Residential Variation Statistics is added in NUTS-1, from the original data for NUTS-2. The correlation between both sources is very high, with values exceeding 95 per cent in most cases and never lower than 82 per cent. Each generic element of the micro-data matrix is denoted by mij/k (t), where i is the region of origin (17 autonomous regions, 2 autonomous cities and abroad), j denotes the destination region (17 autonomous regions and 2 autonomous cities), k the education level (illiterate, without primary education, primary education, secondary education and higher education) and t the specific year of the survey (from 2000 to 2017). These matrixes reflect an inter-regional migratory structure, but the absolute data is limited to the interviewed sample, which requires an elevation adjustment using the data of the “Labour Force Survey subsample: Persons that have changed residence a year ago. Persons ages 16 and over that have changed residence a year ago due to education level attained and place of origin/destination,” whose results are encompassed in 8 NUTS-1 for the places of origin and 7 for those of destination. This elevation adjustment enables us to calculate the ri/k (t) and sj/k (t) margins, which will make it possible to initiate the iterative proportional fitting procedure in order to obtain the 18 annual matrixes of the series (2000-2017).
After adjusting the margins of the micro-data matrix to the absolute levels of the Labour Force Survey subsample, the iterative proportional fitting procedure can be used to calculate the intermediate elements of this matrix, making use of the previous margins. The proposed methodology enables us to attain convergence in the eighteen estimated annual matrixes.
The obtained results favour a broader vision of the relations among regions, with regard to this type of migrations, and enable subsequent univariant or multivariant analyses. They could also be examined using input-output methodology or even under the theory of networks, as proposed in the last section of this paper along with the conclusions.
En el presente trabajo se analiza la robustez estadística de los censos de población y de la Encuesta de Población Activa con el objetivo de estimar matrices regionales de flujos migratorios clasificados por niveles educativos, mediante variantes del iterative proportional fitting, a partir de la información contenida en ellos.
Al no permitir las estadísticas oficiales sobre migraciones la clasificación de los migrantes en función de su formación, es necesario utilizar fuentes alternativas para estudiar los movimientos poblacionales entre comunidades autónomas atendiendo a los niveles formativos.
El recurso a estas fuentes obliga a realizar un análisis previo de las mismas para la elección de la que permita la mejor estimación de flujos migratorios interregionales para los diferentes niveles educativos.
Finalmente se proyectan estos flujos migratorios anuales para el período 2000-2017, mediante técnicas basadas en el iterative proportional fitting.