European Social Survey ESS 2010 Documentation of the Spanish sampling procedure

Transcription

1 European Social Survey ESS 2010 Documentation of the Spanish sampling procedure The 2010 design is quite similar to the ones in 2006 and 2008 with small improvements already introduced in the fourth round. These improvements were: a) A more precise stratification using habitat brackets; b) A more precise estimation of the design effects and c) Elimination of the oversampling in some strata, since the analysis of 2008 response data file did not give any evidence in favour of the necessity of any oversampling. Additionally, in order to achieve a more accurate sample size for the 2010 round, this time the estimations about ineligible rate, response rate and design effect will be obtained as a mean of the results of the last two editions: 2006 and 2008 rounds. This time, any complementary sample will be included in the official ESS data (the 2008 sample included extra samples for Catalonia and Galicia regions). A. TARGET POPULATION The population is composed of all people aged 15 and over residing within private households in Spain including, Ceuta and Melilla, regardless of their nationality, citizenship or language. B. SAMPLING FRAME 1

2 The sampling frame for the 2010 ESS sample is the Spanish population census structured in sections taken from the Continuous Census (Padrón Contínuo) updated in May 2010 by the Instituto Nacional de Estadística (INE, the Public Statistics Office of Spain). Taking the Continuous Census as a sampling frame ensures the best available coverage of the population of residents. In fact, the Continuous Census is updated, using the municipal register: When a citizen moves from one municipality to another s/he has to notify the local authorities of the new place of residence. That will grant her/him access to the public health services, public schools and other public services, as well as the updating the electoral register. The law obliges every Spanish city council to send the data from its register to the INE once a year. This process makes up the national Continuous Census of inhabitants. Foreigners usually register themselves in municipal rolls in order to benefit from welfare services even if they are not legal residents in the country. C. SAMPLE DESIGN The proposed design for the 2010 round of the ESS is a stratified two-stage sample design. The strata are obtained by crossing two population classification criteria. The first criterion is the Autonomous Community or region of residence (there are 17 of them plus another one grouping the North-African autonomous cities of Ceuta and Melilla). The second criterion (the type of habitat criterion) distinguishes among four types of habitat according to their size: The first bracket: cities with more than inhabitants aged 15 and over The second bracket: cities between 50,001 and inhabitants aged 15 and over 2

3 The third bracket: municipalities between 10,001 and 50,000 inhabitants aged 15 and over The fourth bracket: municipalities with less than 10,001 inhabitants aged 15 and over The analysis of 2008 data confirmed the existence of different response rates among habitat brackets, justifying the maintenance of this stratification. More details about the rational behind the proposed stratification will be found in Appendix 1. The cross-tabulation of the two criteria gives a total of 72 theoretical strata (18x4), only 64 of them being effective. In each stratum the two sampling stages are the following: 1. In the first stage, a fixed number of census sections are drawn with probability proportional to the number of inhabitants of 15 years old and over in each section. Thus, census sections are the primary sampling units (PSUs) In the second stage, for each PSU selected in the previous stage, 6 or 7 individuals per unit will be randomly drawn: 7 in the sections belonging to the first two brackets and 6 in the others. As we mentioned above, the data analysis of previous rounds showed a response rate in the two first brackets lower than in the rest (63.1% and 71.1%, respectively, in 2008, and 61.5% and 72.4% respectively, in 2006). The average response rate for the first two brackets is 62.3% and for the last two brackets is 71.7%. The probabilities of inclusion of sections and individuals are provided by the INE. 1 There are 34,600 census sections in Spain. Census sections are the most elementary framing units of eligible voters. The size of sections vary between 500 and 2,000 voters (18+ years old), being the average size of 1,300. Nevertheless, it should be stressed that although census sections are defined with regard to electoral processes, these are only used for establishing the boundaries of administrative units that are used for sample designs. Census sections do include all citizens registered in the municipal registers, regardless of their voting rights. 3

4 D. DESIGN EFFECTS The effect due to different probability of selection is not included in the design effect of the Spanish survey because the selected sample is a sample of individuals instead of a sample of households. As a result, the total design effect (DEFF) is equal to the design effect due to clustering (DEFF c ). For the estimation of the effect of clustering in the 2010 round, 22 variables of 2006 round and 23 variables of 2008 round have been used. See Appendix 2 for the calculations leading to the final estimation of the 2010 mean design effect: DEFF= DEFF c =1.207 E. RESPONSE RATE The first two ESS rounds highlighted the difficulties for achieving the target response rate of 70% in Spain. The response rates were 53% in 2002 and 55% in However, due to the improvements implemented in the fieldwork plan and the serious involvement of the survey company in the way that they conducted and controlled the fieldwork, the Spanish response rate raised to 66.7% in 2006 and to 66.8% in The goal of 70 % is within reach. In the calculation of the 2010 sample size an estimated response rate of 66.8% has been used. F. VALID CASES 4

5 The proportion of valid cases in the 2008 Round was 0.973, considerably higher than in the previous round (0.870) and closer to the 2004 round (0.950). Taking into account that the population data used in the 2010 sampling design have been recently updated (May 2010), we preview a high eligible rate for 2010 round. Due to all, we estimate the 2010 eligible rate as a weighted average of the two last eligible rates assigning a higher weight for the last one. Estimated proportion of valid cases in 2010 Round = 0.973*(2/3) *(1/3) = G. SAMPLE SIZE In the calculations of the sample size three numbers have to be estimated: the proportion of valid cases, the response rate and the design effect. As we have explained in previous paragraphs, estimations for the 2010 design have been obtained from the two previous rounds. The values for these estimations are: Proportion of valid cases = Mean response rate = Design effect = Taking into account the above information, the calculations to determine the sample size for the 2010 survey are the following: Minimum effective sample size = 1,500 Net sample size = 1, =1,811 Gross sample size = 1,811/( ) = 2,886 In the process of assigning individuals to each stratum proportionally to the population of inhabitants of 15 years old and over, the constraint to take an 5

6 integer number of sections in each stratum has led to a slight modification of the total number. Therefore, the total sample size is 2,865. Table 1 presents the distribution of the number of sections and individuals to be selected in each stratum proportional to the population of inhabitants of 15 years old and over. See Appendix 3 for the distribution of the 2010 Spanish population of inhabitants of 15 years old and over. Table 1. Distribution of sections and individuals by strata (proportionally to the population) Number of sections Number of individuals Size of habitat Region More than 50,001 and 10,001 and 50,000 Less than 10,001 Total More than 50,001 and 10,001 and 50,000 Less than 10,001 Total Andalucía Aragón Asturias Baleares Canarias Cantabria Castilla y León Castilla-La Mancha Cataluña Valencia Extremadura Galicia Madrid Murcia Navarra País Vasco La Rioja Ceuta y Melilla Total

7 Appendix 1: Stratification We discuss below the reasons for stratification by region and type of habitat: Stratification by region. There is a common practice for social surveys in Spain to use stratification by Autonomous Communities (regions). That procedure is based on the observed socio-economic, political and cultural differences. The analysis of the four first rounds ESS data corroborated those differences among regions for some variables and, thus, the benefit to stratify by autonomous communities although the mean design effect due to stratification seems to be negligible. Stratification by type of habitat. From the first stratification in the 2002 Round there has been an improvement in the stratification by type of habitat. Response rates by habitat brackets justify the applied stratification: the bigger the city the lower the response rate. Results of the second round also suggested the need to reconsider the stratification with an aim to reducing the heterogeneity in terms of town size within the strata. Thus, from the third round the stratification has been composed of the following four brackets: cities with more than inhabitants, cities between 50,001 and inhabitants, municipalities between 10,001 and 50,000 inhabitants and, finally, municipalities with less than 10,001 inhabitants. Table 2. Expected and observed response rate (%) and cluster s size per bracket for the 2008 round Size of habitat: People aged 15 years and over More than 50,001 and 10,001 and 50,000 Less than 10,001 Total 2008 previsions survey data Percentage of 2008 population Individuals by section

8 Table 3. Expected response rate (%) and size of cluster per bracket for the 2010 round Size of habitat: People aged 15 years and over More than 50,001 and 10,001 and 50,000 Less than 10,001 Total 2010 previsions Percentage of 2010 population Individuals by section

9 Appendix 2: Design effects estimation The formula used in the calculation of design effect due to clustering is the following: DEFF c = Being: 1+ ( k 1) ρ ρ = intra-group correlation coefficient k = estimated average number of completed interviews per cluster To estimate the intra-group correlation coefficient in the 2010 round we have used the data from 2006 and 2008 rounds for a group of numerical, ordinal and dummy variables. All of the variables except one were also used in the 2008 design. Some of the ordinal variables were also used in 2006 design and others were proposed by the ESS experts panel. En effect, for the 2010 calculations, we follow the recommendations included in the document Sampling for the European Social Survey Round V: Short version. In that document the use of two additional variables not included in the calculus of 2008 design are suggested. These are dscrrce and dscrntn. Unfortunately, these two variables have a very low response rate in the 2008 Spanish data making them not suitable for the estimation. Instead of dscrrce and dscrntn another related more general variable named dscrgrp has been selected. Table 4 provides the prevision for the average number k of completed interviews per cluster while Table 5 displays the list of selected variables and the calculation of the value of the intra-group correlation coefficient used in the estimation of the design effect of clustering. The intra-group correlation coefficients (ρ) for each of the variables have been estimated using two level variance decomposition models. Each PSU is considered as a group or cluster (level-2 unit). 9

10 We computed the means of the number of completed interviews for cluster in 2006 and 2008 rounds. The prevision for k is the average of these two means. Table 4. Estimation of the mean response rate per cluster 2006 data 2008 data Mean (2010 prevision) k Table 5. Intra-group correlation coefficient Variable 2006 ρ 2008 ρ Mean ρ Ordinal PPLTRST PPLFAIR PPLHLP POLINTR TRSTLGL TRSTPLT STFECO STFGOV STFDEM DSCRGRP Numerical HHMMB YRBRN EDUYRS PDJOBYR WKHCT Dummy VOTE PDJOBEV MOCNTR GNDR UEMP3M UEMP12M UEMP5YR CHLDHHE Total Finally, the design effect due to clustering for 2010 round is estimated by: 1 + ( k 1) ρ = 1+ ( ) =

11 Appendix 3: Assignment of the number of individuals and sections to strata Table 6 shows the distribution of the Spanish population in the 64 strata considered: Table Spanish population of 15 years old and over per strata Size of habitat: cities with Region More than 50,001 and 10,001 and 50,000 Less than 10,001 Total Andalucía 2,235,132 1,154,560 1,961,431 1,602,891 6,954,014 Aragón 582, , ,736 1,166,075 Asturias 447,402 74, , , ,824 Baleares 342, , , ,354 Canarias 649, , , ,838 1,792,631 Cantabria 161, , , ,897 Castilla y León 690, , ,477 1,045,887 2,262,265 Castilla - La Mancha 142, , , ,070 1,764,284 Catalonia 2,629, ,512 1,643,174 1,315,302 6,355,569 Comunidad Valenciana 1,334, ,389 1,619, ,181 4,343,501 Extremadura 123,827 79, , , ,406 Galicia 477, , , ,224 2,477,980 Madrid 3,797, , , ,301 5,433,151 Murcia 537, , ,046 81,426 1,195,436 Navarra 171, , , ,942 País Vasco 680, , , ,473 1,890,447 Rioja (La) 129, , , ,367 Ceuta y Melilla 0 119, ,784 Total 15,133,184 4,796,005 10,536,091 9,462,647 39,927,927 Source: INE, 2010 Continuos Census The total number of sections to be selected comes from the gross sample (2,886) divided by the number of individuals per section (6.5), giving an initial total of 444 sections. This total has been distributed among the four brackets of habitat proportionally to their population of 15 and over. The assignment of 7 individuals per section in the first two brackets and 6 in the rest gives the final distribution of individuals to be selected in each bracket as it has been provided in Table 1. 11