quota and stratified sampling.
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
doi:10.5477/cis/reis.171.23
Household Sampling Designs: Differences and Similarities between Probability
Sampling and Route and Quota Sampling Diseños muestrales en hogares: diferencias y similitudes entre muestras
probabilísticas y muestras con rutas y cuotas
Vidal Díaz de Rada and Valentín Martínez
Key words Quota Sampling
Random Route • Method
Random Sampling• Multistage Cluster •
Sampling Sampling Methods•
Abstract This paper compares the representation quality of three face-to- face household surveys. Two of them used probability samples and the other one selected the ultimate sampling units by using random route and quota sampling, with non-responses resulting in ‘automatic’ substitutions. The hypothesis to be tested is that random route sampling and quota sampling (with substitution) provide similar representative quality as home sampling (without substitution) based on the local population register. Marked differences were found in education level in the probability samples, where the deviations exceeded 25%. A different picture emerged when comparing employment variables, where quota sampling overestimated both the labour force participation rate (by 2.5% points) and unemployment rates (9.5% points).
Palabras clave Método de cuotas
Método de rutas• Muestreo aleatorio• Muestreo por •
conglomerados en varias etapas
Selección muestral•
Resumen Este artículo compara la representatividad lograda en tres encuestas presenciales en el hogar. Dos emplean muestreos probabilísticos y la tercera una selección de las unidades últimas mediante un sistema de rutas y cuotas, llevando a cabo sustituciones «automáticas» cuando no se consigue una respuesta. Se busca contrastar la hipótesis de que la representatividad lograda por un muestreo por rutas y cuotas (con sustitución) es similar a la conseguida en muestreo de viviendas (sin sustitución) basado en el Padrón. Los resultados muestran grandes diferencias en el nivel educativo mostrado por las muestras probabilísticas, con desviaciones superiores al 25%. Los resultados son diferentes en las variables de empleo, donde las encuestas con cuotas sobreestiman las tasas de actividad (en 2,5 puntos porcentuales) y paro (en 9,5 puntos porcentuales).
Citation Díaz de Rada, Vidal and Martínez, Valentín (2020). “Household Sampling Designs: Differences and Similarities Between Probability Sampling and Route and Quota Sampling”. Revista Española de Investigaciones Sociológicas, 171: 23-42. (http://dx.doi.org/10.5477/cis/reis.171.23)
Vidal Díaz de Rada: Universidad Pública de Navarra | vidal@unavarra.es
Valentín Martínez: Centro de Investigaciones Sociológicas | valentin.martinez@cis.es
24 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
IntroductIon1
According to the specialist literature, household probability sampling is the only method that allows the results to be extrapolated to the universe. However, it poses numerous difficulties, due to the vast amount of information necessary for its correct application. This is clearly one of the factors that has given rise to a pro- liferation of studies that have proposed using non-probability sampling (among others, Mercer et al., 2017; Miller, 2017).
In Spain, most of the household sur- veys conducted by the private sector on market and opinion research use sam- ples based on random routes and quotas (among others, Núñez Villuendas, 2005; Cuxart and Riba, 2009; Alvira, 2011) that employ non-probability selection crite- ria. However, Spain’s participation in in- ternational research projects (Euroba- rometer, Comparative Study of Electoral Systems, World Values Survey, European Social Survey, International Social Sur- vey Programme, etc.) that demand identi- cal methodologies in all countries (among others, Smith, 2010), together with the harmonisation of official statistics caused by having to conform to the rules im- posed by Eurostat, has proven to be a challenge for Spanish research. One of the basic requirements of these studies is the use of probability sampling. But this is difficult in Spain, as there has not been much of a tradition for using this type of
1 This study was carried out during a research visit to the Sociology Department of the University of Nebras- ka-Lincoln (UNL), funded by the Public University of Navarra. The author wishes to thank the Department for their warm welcome, in particular, Julia McQuillan and Jolene D. Smyth.
This text is part of a study funded by the Spanish Ministry of Economy and Competitiveness, reference CSO2012-34257. The author wishes to thank the Edi- torial Board and the two anonymous reviewers for their suggestions, which improved the original version of the paper.
sampling–except for the National Institute of Statistics.
This article compares the representa- tive quality achieved in three face-to-face household surveys. Two of them were conducted using probability sampling and the other one using a random route method, and cross-over age and sex quo- tas to select the ultimate sampling units. Non-responses from the sample unit re- sulted in “automatic” substitutions. The aim is to evaluate the information from six variables and to verify the hypothesis that the representative quality achieved by routes and quota sampling (with substi- tution) is similar to that obtained in home sampling that reduces non-response rates by using a revisit instead of replacement. This hypothesis gives rise to several sub- hypotheses:
— H1: Route and quota sampling (age and sex) achieves better representa- tion quality of the universe distribution by age and sex when these character- istics are considered by interviewers in respondent selection.
— H2: There are likely to be more highly educated respondents when using probability sampling, since various studies (Beullens et al., 2018; Wil- liams and Brick, 2018; National Re- search Council, 2013, among others) have found a relationship between in- creased educational level and cooper- ation rate.
— H3: The use of revisits in probabil- ity sampling selects a greater number of employed persons (outside of the household). The use of substitutions in the sample with routes and quotas generates an over-representation of the unemployment rate.
— H4: The working day of employees, which is shorter than that of employers and self-employed people, means that
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 25
employees are better represented in route and quota sampling.
The study is organised into 5 parts. Af- ter explaining the purpose of the study, the basic criteria of probability sampling and of random route and quota sampling for the selection of the ultimate respond- ents is briefly described. The sources of data used and the methodology for car- rying out comparisons between distribu- tions are then outlined. The fourth sec- tion includes comparisons between the three studies, considering the distribution by age and sex, educational level, and a comparison of the three employment vari- ables.
ProbabIlIty SamPlIng
T h e s p e c i a l i s e d l i t e r a t u r e d e f i n e s p r o b a b i l i t y s a m p l i n g a s a m e t h o d i n which all the units have a known prob- ability of being included in the sample, and which employs a selection criterion that respects this probability. A frame is therefore needed that includes all of the members of the universe, that is, an ex- haustive list of all sampling units without duplication (Scheaffer et al., 2007).
I n S p a i n , t h e N a t i o n a l S t a t i s t i c s Institute (known as INE by its Span- ish initials) is the body responsible for coordinating the Local Population Reg- ister -Continuous Register with the Lo- cal Councils, as well as for preparing the Electoral Register. All these opera- tions are continuously updated. In this way, the INE has an updated list of all the people who reside in Spain Span- ish, as well as the necessary information to locate individuals, including their full name and address.
This is the basis of the selections used in two of the surveys used in this study, the European Social Survey and the Role
of Government (III)/Work orientations (I) study, both carried out by the Centre for Sociological Research (known as CIS for its initials in Spanish). The CIS requested the details of the individuals’ who took part in both surveys, taking into account the sample design characteristics of each of them.
SamPleS baSed on IncomPlete frameS
A s n o t e d a b o v e , t h e i n f o r m a t i o n needed to conduct household sampling is only accessible to a limited number of bodies2, which leads the private sector or- ganisations engaged in market and opin- ion research to use other strategies.
Of all the existing strategies for lo- cating people to interview, one of the most used is the random route sampling method (among others, Bréchon, 2015; Bauer, 2016), which selects the house- holds through a route that the interviewer must follow to carry out the assigned in- terviews. This includes randomly cho- sen streets that prevent interviewers from deciding the route to follow, so they are “forced” to take certain streets (Díaz de Rada, 2015). Each route has a starting point, usually a specific address, and from there, a series of criteria must be followed for the selection of homes. Five different routes are presented in the sec- ond chapter of Díaz de Rada’s publica- tion (2015).
Once the homes have been selected, the next step is to choose a person within the household, provided it is not a single- person household, as is the case in 75%
2 In fact, the volume of surveys conducted in Spain and the costs of these processes make it difficult to use this service for the entire research sector through surveys.
26 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
of Spanish homes (INE, 2016b)3. In Spain, most of the people residing in households (ultimate units) are selected by the so- called quota method, which tries to obtain a sample that is similar to the population under study based on certain characteris- tics. For this, a series of characteristics of the units to be interviewed are selected, generally referring to various sociodemo- graphic features such as age, sex, em- ployment relationship status, profession, educational level, etc., and subsequently a series of “selection sheets” are pre- pared, which list the characteristics of the respondents. This method is based on the premise that a sample that is similar to the population in a series of important characteristics will also be similar in re- spect to the other characteristics of inter- est for the analysis.
Quota sampling has the advantages of simplifying the fieldwork and significantly reducing the costs involved. Sudman, for example, estimated that quota sampling is three times cheaper than probability sam- pling with more than one household visit (Sudman, 1976: 199). The main disadvan- tage is the high degree of freedom granted to the interviewer, which generates signifi- cant biases in the selection process (Me- nold, 2018) that are difficult to detect. The following paragraphs will explain in detail what this greater “freedom of the inter- viewer” involves.
3 The authors are fully aware of this change in ter- minology from home to household. A Home, accord- ing to the INE (2016a), is “a structurally separate and independent enclosure which, due to the way in which it was built, rebuilt, transformed or adapted, is designed to be inhabited by people or, even if it is not, it is someone’s habitual residence at the time of the survey” (INE, 2016a: 3). In contrast, a house- hold refers to living situations: “the person or group of people who usually reside in a main family home” (INE, 2016a). Therefore, there can be more than one household in a home, although this is uncommon. The change in terminology is due to the fact that the object of study of most of the surveys is the house- hold.
The fundamental difference between the quota method and other methods (Kish selection method, birthday-based meth- ods, etc.) is the way in which people are selected to be part of the sample (Mar- lar et al., 2018). Thus, for example, if in a given household the interviewer finds two women aged between 16 and 30 years old, the interviewer decides how to se- lect which person to interview. This over- rides the maxim of probability sampling that postulates that all the units of the population must have the same probability of being included in the sample (Menold, 2014). When the interviewer begins the route and goes to the first household, s/he can choose any of the people included in the selection sheet, and s/he chooses the person to interview. When the person cho- sen cannot be interviewed (or refuses to cooperate) the interviewer does not try to persuade the individual to collaborate, as in probability methods (Butcher, 1995), but instead goes on to choose another person within the household, provided that they comply with the characteristics laid out on the “selection sheet”.
In the first interviews on the route, the sheet is “intact”, so obtaining an interview in the first household visited is very easy. Each time an interview is completed, the characteristics of the interviewed person are crossed out. This means that, by the end of the route, the interviewer has to search for people with very specific charac- teristics, who are more difficult to find (Díaz de Rada, 2008). Both situations mean that the interviewer has the freedom to choose the person to interview.
In the case being discussed here, since there are several people in the household who could be interviewed, finding a re- placement does not pose problems for the interviewer. But what happens when no- body responds, or when the selected per- son refuses to cooperate? Or when nobody in the household meets the required char-
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 27
acteristics? When people to be interviewed cannot be identified when using quota sam- pling, and there are downright refusals to participate, or any other incident that pre- vents a person from being interviewed, a substitution is made. This entails adding new items to the sample that replace the selected people who did not respond (Me- nold, 2014). This is one of the most com- monly used strategies at present, mainly by the private sector in opinion and market re- search (Rotman and Mitchell, 1989; Sud- man and Blair, 1999; Taylor et al., 1995; Sánchez Carrión, 2000), as it provides a rapid “solution” to the non-response prob- lem and makes it possible to obtain a previ- ously fixed sample size.
The strategy to be used depends on the researcher, and in the route and quota sampling used in this study, the procedure is set forth in the document entitled General rules for the correct application of the sam- ple: “when an interview is not conducted at the time of the first contact, try again at the next door” (CIS, 2011). In this way the next interview is carried out at the first door of the next segment (6-home group). In the case of entrance foyers to blocks of flats where no interview can be made “the selected entrance foyer is replaced by the next one” (ibid.).
At first, this causes numerous unsuc- cessful calls. Núñez Villuendas (2005) used 88 Barometers carried out by the CIS between 1996 and 2003 and noted out that this problem affected 27% of visits to homes. She noted that “there are quotas which need three hours to complete the sample sheet” (Núñez Villuendas, 2005: 5). The second consequence of making sub- stitutions is the introduction of biases in the sample, by collecting information from units other than those originally chosen. Thus, for example, in the Living and Work- ing Conditions Survey in Spain, Murgui et al. (1992) reported that the use of substitution produced greater over-representation of
women over the age of 65, compared to the low representation of men aged between 20 and 24 years old. The problem was not limited only to sociodemographic features, but it also affected the specific aspects of the study. Bréchon (2015) pointed out that when a person who refuses to participate is replaced by a neighbour (who then agrees to cooperate), the latter usually has greater social involvement, and a greater participa- tion in social and political life. This results in units being introduced that are different from those they replace.
Some experiments that compared the selection made by this method with other probability methods did not find signifi- cant differences (among others, Rodríguez Osuna, 1991; Bréchon, 2015), although most of the existing literature criticises this procedure, because the quota method in- cludes a small number of rejections and people who are difficult to locate (among others, Worcester and Downham, 1986; Marsh and Scarbrough, 1990). However, the industry argues that, while these situ- ations must be kept in mind, the samples “work”, and sometimes better results are obtained than those provided by strictly random sampling (among others, Sudman and Blair, 1999; Bréchon, 2015).
data and methodS
The studies used for comparison with probability sampling were the 8th round of the European Social Survey and Role o f G o v e r n m e n t / O r i e n t a t i o n s t o w a r d s work, both conducted by the CIS (Surveys No. 3167 and 3135, respectively). These were compared with the Health Barome- ter of the second semester of 2016, which used a system of random routes to select the homes, and quotas for within-house- hold respondent selection. The methodo- logical designs of each of these studies are described below.
28 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Data Sources
Probability surveys
The object of study of the European Social Survey (hereinafter ESS) is the pop- ulation aged 15 and over residing in main family households in Spain (Cuxart and Riba, 2009), while the study focused on the Role of Government/Orientations to- wards work within the International Social Survey Programme (hereinafter ISSP) con- centrated on the over 18 years old popu- lation (Centre for Sociological Research, 2016a). Both used the Local Population Register from January 2015 (European So- cial Survey, 2018a; Centre for Sociological Research, 2016a)4 as a sampling frame. The universe was stratified by autono- mous region and habitat, considering four categories in the ESS and seven in the ISSP. Both surveys used the same units (sections and individuals) and the same sampling procedure: two-stage with se- lection of sections of the population reg- ister, with probability proportional to their size, followed by a systematic selection of the individuals in the section, once they had been ordered by the number of homes where they resided. The initial sample size of both samples was around 3,000 inter- views (3,080 in the ESS and 3,000 in the ISSP) and employed a proportional alloca- tion in the defined strata.
The ESS fieldwork began on 16 Febru- ary and ended on 26 June 2017, while the ISSP was conducted between 11 April and 29 June 2016.
Another characteristic element of these surveys was that the population who had been selected received a letter of intro- duction prior to the interviewer’s visit, where they were told that they had been selected to participate in a survey, the ob-
4 All the information included in that section was taken from both sources, except when other references are cited.
jectives of the survey were explained, and they were informed that there was a free telephone number and an email address they could contact if they wanted further information. In addition, both used a bro- chure that briefly explained the purpose of the survey, and how the answers were used.
In the case of ESS, those who refused to cooperate received another letter that emphasised the importance of their par- ticipation in order to obtain suitable rep- resentation. This letter that included a free telephone number, and an e-mail address they could use to contact the person who was conducting the survey. In addition, in recognition of their effort, after answering the questionnaire the respondents would receive EUR 9 in the ESS (in the form of a purchase voucher), and a tote bag with the CIS logo in the ISSP.
In both studies, the impossibility of re- placing individuals who refused to co- operate necessitated at least four visits to homes where contact was not estab- lished, which were carried out at different times and at least one of them during the weekend. The ESS also used “refusal con- version” strategies.
All these resources achieved the com- pletion of 1,958 questionnaires in the ESS, and 1,834 in the ISSP, which represented cooperation rates of 64.4% and 61.3%, respectively (COOP1 American Associa- tion for Public Opinion Research-AAPOR, 2016: 63), with a total of 3,038 and 3,000 contacts made, respectively.
Study by random routes and quotas
The universe of the 2016 Health Ba- rometer was the resident population aged 18 and over, stratified by autonomous re- gions and 7 habitat categories, the same as the ISSP study (CIS, 2016c and 2016d). The study consisted of three representa- tive subsamples of the Spanish popula-
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 29
tion, although the second one was con- sidered in this comparison. The field work was carried out between 10-19 June 2016. It used multi-stage cluster sampling in which the primary units (253 municipali- ties) and the secondary units (population register sections) were selected on a pro- portional, random basis (Martínez Martín, 2004). The design started with a uniform distribution of 250 interviews in each com- munity and was extended proportionally with a number of interviews to reach a to- tal of 7,800 interviews. Each wave of the Barometer was one third of the total sam- ple, that is, 2,600 interviews (CIS, 2016c). the second wave of the Health Barome- ter from 2016 (CIS Survey No. 3140) was chosen due to the similarity with the prob- ability method in the sample frame, in the population under study, and in the dates when fieldwork took place.
The non-proportional sampling adopted meant that the regional results could be closely compared. The objective was to discover and compare people’s views on different aspects of the health system by autonomous region. This implied that the more sparsely populated communities were over-represented, while those that were more densely populated were under- represented, due to the distribution of the resident population. This caused an over- representation of smaller regions (such as La Rioja, Navarra, etc.) and an under- representation of larger regions. Conse- quently, in order to obtain representative data at the national level, it was necessary to make some adjustments and increase the weight of the most populated autono- mous regions and decrease the weight of the least populated (CIS, 2016c).
The homes were chosen by using a system of random routes within the pop- ulation register section selecting the ulti- mate units (individuals resident in those homes) using related sex and age quo- tas (Martínez Martín, 2004), with six age
groups: 18-24 years old, 25-34 years old, 35-44 years old, 45-54 years old, 55-64 years old and, and 65 years old and above.
Whenever the interview could not be carried out (no response, refusal, etc.), the sample unit was replaced following the instructions established in the Gen- eral rules for the correct application of the sample document discussed earlier. This procedure meant that in order to com- plete 2,587 interviews it was necessary to contact 62,044 homes, with an average of 23.9 contacts per interview. More than half of these contacts (55.5%) were from homes where no one responded, 10% from homes where people refused to co- operate, and another 6.8% from homes where residents refused before explaining that it was a survey. On 2.9% of the occa- sions it was not possible to gain access to the building (house, housing estate, etc.), and 3.15% of the contacts were made in places that were not homes (offices, doc- tor’s surgeries, etc.). In addition, 20.5% of the contacts did not result in an interview because the quotas had already been cov- ered.
In short, the three studies performed a similar stratification, depending on the size of the municipality in question, and used cluster sampling where the ultimate cluster was the population register sec- tion (direct selection of sections in the case of the ESS and ISSP, and selec- tion of municipalities and sections within the selected municipalities in the case of the Health Barometer). The differences between the three studies began at this point, since the Health Barometer used routes and quotas, and probability sam- pling studies employed lists of people. The ESS used more resources to increase collaboration: an additional cover letter (compared with the ISSP), refusal conver- sion, and a monetary reward. A money in- centive has been reported by almost all
30 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
researchers on the subject to be more ef- fective than the use of any gifts (among others, Ernst Stähli and Joye, 2016). An- other essential difference was that the Health Barometer included the replace- ment of non-located units, while the other two did not. Regarding the inequality in the ESS universe, which considered peo- ple aged 15 and older, those respondents under 18 were eliminated in order to fa- cilitate the comparison, which resulted in the sample size being reduced to 1,818 respondents.
Analysis: Comparison between distributions
Representative quality was studied by comparing the distribution achieved by each survey and the reference population (Martínez Martín, 2004) in the variables for which there was information from the population as a whole, specifically the dis- tribution by sex and age, provided by the Register on 1 January 2016 in the ISSP and in the Health Barometer, and 2017 in the case of the ESS.
The study would lack interest and originality if it were restricted to that comparison, so it was extended to other variables, including educational level and labour relationship. Although there was no updated information on the universe, the Spanish Labour Force Survey (here- inafter LFS) provided a good approxima- tion to the Spanish population as a whole (Díaz de Rada and Núñez Villuendas, 2008). This is the main survey targeting households, taken into account sample size, cost, and personnel used. In fact, the INE website reported that the cost of the 2019 LFS was 11.408 million Euros (INE, 2019).
To find the representative quality of each survey, the joint distribution of each variable of interest and sex was consid-
ered. Both variables were considered to- gether to identify deviations which would be hidden in certain frequencies if the marginals were used, since the compen- sation employed between subgroups could conceal this deviation. For exam- ple, in the second part of Table 1, the marginal distribution of the 55-64 year- old group was over-represented by 0.59 points when compared with the popula- tion register data—something that could be indicative of a good fit. However, when disaggregated by sex, there was an un- derrepresentation by nearly 1 point in the case of men (0.85), and an over-repre- sentation by 1.44 points in the case of women. This involved an aggregate mis- match of 2.29 points. The women in this age group were therefore identified as being “responsible” for the deviation de- tected.
The distribution of each (sample) vari- able will be shown in the tables of the fol- lowing section compared with the corre- sponding information from the universe. Thus, as can be seen in Table 1, the sec- ond column contains the age distribution of men under the term “data” and, on the right, the differences when compared with the Register. The calculation, subtracting the distribution of the universe from the sample distribution, meant that the posi- tive values meant a sample over-repre- sentation (in this case of men aged 65 and over), and the negative figures meant an under-representation with respect to the universe.
The sum of the differences (SD) was added to the distribution considering each cell at the end of each table, show- ing the different fit of men and women, and the sum of absolute deviations (SAD). This last sample shows the total magni- tude of deviations for each distribution, which provides values higher to the first because the SD compensated for the dif- ferences.
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 31
reSultS: dIfferenceS between the SamPlIng ProceSSeS
Distribution by age and sex
Table 1 details the age distribution achieved by each study compared to the reference universe. The comparison reveals a total difference of 13.6 points, mainly caused by the under-representation of the group between 25 and 44 years old, as well as the over-representation of the group of that age and above. A slight over-represen- tation of the under 24-year-old group was also detected. The sex-differentiated anal- ysis revealed a slightly poorer representa- tion of men, with an SAD value of 7.43 (6.16 in women). Males between 25 and 44 years
old were under-represented by 3.19 points, while the opposite happened for the 45-55 age group, although to a lesser extent. In the case of women, the youngest appeared to be over-represented by 1.14 points, and the opposite occurred in the 25-34-year-old age group.
Dividing each magnitude by the to- tal differences (13.60) revealed the sub- groups that contributed most to the de- viations detected. In this case, 13.9% of the differences occurred in men over 64, and 13.8% among men between 25 and 34 years old. The next highest magnitude (12.25%) corresponded to women be- tween 25 and 34 years old. These three subgroups amounted to 39.9% of the dif- ferences.
TABLE 1. Comparison between the sample and the universe in the distribution of age and sex. Vertical percent- ages and differences between magnitudes (sample minus universe5)
European Social Survey 2017 (Round 8) (CIS Survey 3167, year 2017)
Men Women Total
Data Difference Data Difference Data Difference SAD
18-24 4.401 0.152 5.20 1.14 9.60 1.29 1.29
25-34 5.40 –1.87 5.60 –1.66 11.00 –3.54 3.54
35-44 9.00 –1.32 9.20 –0.85 18.20 –2.17 2.17
45-54 10.80 1.12 10.10 0.50 20.90 1.62 1.62
55-64 8.60 1.08 8.80 0.96 17.40 2.05 2.05
65 and over 11.40 1.89 11.60 –1.04 23.00 0.84 2.93
Total 49.60 50.50 100.10
SD 1.05 –0.95 0.10
SAD 7.43 6.16 11.51 13.60
1 Percentage of men between 18 and 24 years of age compared to the total number of respondents in the ESS.
2 This value was obtained by subtracting the universe distribution from the sample distribution, with the negative values repre- senting an under-represented sample, and the positive numbers an over-represented sample with respect to the universe.
5 The difference between the values obtained from the sample and the population were interpreted as indicating over-representation if they were positive, and under-representation if they were negative.
32 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
TABLE 1. Comparison between the sample and the universe in the distribution of age and sex. Vertical percent- ages and differences between magnitudes (sample minus universe) (Continuation)
Role of Government/Work orientations (CIS Survey 3135, year 2016)
Men Women Total
Data Difference Data Diferencia Data Difference SAD
18-24 4.20 –0.07 3.80 –0.29 8.00 –0.36 0.36
25-34 6.50 –1.04 6.70 –0.82 13.20 –1.87 1.87
35-44 9.50 –1.01 10.90 0.73 20.40 –0.28 1.74
45-54 9.30 –0.27 10.50 0.99 19.80 0.72 1.27
55-64 6.50 –0.85 9.10 1.44 15.60 0.59 2.29
65 and over 10.90 1.56 12.00 –0.47 22.90 1.09 2.03
Total 46.90 53.00 99.99
SD –1.68 1.58 –0.10
SAD 4.80 4.75 4.91 9.55
Health Barometer (2nd wave) (CIS Surveys 3133 and 3140, from 2016)
Men Women Total
Data Difference Data Difference Data Difference SAD
18-24 4.30 0.03 4.30 0.21 8.60 0.24 0.24
25-34 7.90 0.36 8.10 0.58 16.00 0.93 0.93
35-44 10.50 –0.01 10.10 –0.07 20.70 –0.08 0.08
45-54 9.30 –0.27 9.10 –0.41 18.40 –0.68 0.68
55-64 6.90 –0.45 7.10 –0.56 14.00 –1.01 1.01
65 and over 9.70 0.36 12.70 0.23 22.40 0.59 0.59
Total 48.60 51.40 100.10
SD 0.02 –0.02 0.00
SAD 1.47 2.05 3.52 3.52
Source: European Social Survey 2018b; Centre for Sociological Research, 2016c and 2016d. Data of the universe taken from the INE: 2017a and 2016c.
The ISSP had a better overall fit and lower SAD values than the ESS. The big- gest mismatch was related to the over- representation of men and women be- tween 55 and 64 years. That group had the worst fit, with absolute differences
(SAD) in excess of two points (2.29) as a result of the smaller number of men and the numerical supremacy of women. Both groups were responsible for 24% of the total deviations. In addition, men were un- der-represented in the under 65 group,
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 33
and under-representation was greater in the 25-44 age group. This under-repre- sentation of men accounted for 50.3% of the differences found.
The comparison with the Health Ba- rometer showed a greater similarity in all age groups, as well as in the distribution by sex. It should be noted that there was lower collaboration rate of the strata be- tween 55 and 64 years old, which was greater in women, as they presented a joint under-representation slightly above one percentage point. The situation was reversed in the oldest group, where there w a s o v e r - r e p r e s e n t a t i o n , w h i c h w a s slightly higher in the male group. These two age groups considered together ac- counted for 45.2% of the overall differ- ence. There was also over-representation of both men and women under 35, al- though this was more pronounced among women.
An overview of the findings showed that the comparison of the age and sex distribution of the sample by routes and quotas achieved smaller differences than those samples that did not use them. The fact that the same variables were used in the selection of the respondents only partially accounted for this fit, since the interviewers were able to “replace a quota with the neighbouring one” when there were difficulties in locating a spe- cific individual6. Although the literature
6 The General rules for the correct application of the sample (Centre for Sociological Research, 2011) pro- vided that if an interviewer found it difficult to locate a respondent, s/he could replace this person with an- other one from the neighbouring age quota. This is the exact wording:
When it is impossible to obtain a certain age quota, it can be substituted by one of the neighbouring quo- tas, although no more than one change can be made per sample sheet. The questionnaire will record the real age and the change made will be recorded on the questionnaire and in the report, where the sam- pling point and reason for the change will also be re- corded.
showed a lower response rate (it should be noted that 23.9 contacts per effective interview were necessary) and a greater difficulty in finding young people (among others, Pasadas del Amo et al., 2006; Díaz de Rada and Núñez, 2008), in this case the biggest differences occurred in those over 45, and more in women than in men.
Differences in educational level
The answers to the educational level questions were re-categorised in order to “match them” to the categories used in the Labour Force Survey, and a greater number of significant differences between universe and sample were found here. As can be seen in the first part of Table 2, the differences in the ESS were mainly caused by the greater under-representa- tion of people with first-stage secondary education and the over-representation of people without any formal education, and those who had completed a Vocational Training course (hereinafter VT). The dif- ferences in secondary education among men and women, divided by the total dif- ferences (61.10%), showed that both rep- resented 39.93% of all the deviations de- tected in the table, which increased to 61.90% when the over-presentation of the respondents with VT education was added. Those who had completed higher education (both VT and higher) were also under-represented (this was more pro- nounced among men).
The ISSP again contained an over- representation of people with VT educa- tion, although to a lesser degree, with similar values for men and women; and it under-represented people whose max- imum educational level was first stage secondary and higher (this under-repre- sentation was more pronounced among men). The subgroups that contributed the
34 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
most to the total deviations were those with VT and first-stage secondary edu- cation, which were responsible for 58% of the differences. These differences in- creased to 72.7% when men with higher education were also considered. It should be noted that the educational level varia- ble is over-represented for women in this survey, mainly due to the smaller number of men with secondary and higher educa- tion. In any case, this was the survey with the best fit of the three analysed.
The Health Barometer (routes and quotas) was the second survey with the greatest differences, eight points above the ISSP fit. These differences repre- sented almost half (31.3) of those found in the ESS. Despite this greater difference, the trends were very similar: respondents
with VT were over-represented, and two groups were under-represented, namely those with secondary first-stage educa- tion, greater difference in women, and those who had completed higher edu- cation, greater under-representation in men. The under-representation of these three subgroups accounted for 40.4% of the deviations from the table, which in- creased to 69.4% when taking into ac- count the over-representation of people whose maximum educational level was Vocational Training.
The Barometer also under-represented respondents without any formal educa- tion, particularly women. This was the main difference with respect to the other two surveys where they were over-repre- sented, especially in the ESS.
TABLE 2. Comparison between the sample and the universe in the distribution of educational level and sex. Ver- tical percentages and differences between magnitudes (sample minus universe)
European Social Survey 2017 (Round 8)
Men Women Total
Data Difference Data Difference Data Difference SAD
No formal education 9.301 6.232 11.90 7.30 21.20 13.53 13.53
Primary 9.30 2.93 8.20 0.62 17.50 3.55 3.55
Secondary (1st stage) 3.00 –12.30 2.60 –11.19 5.60 –23.49 23.49
Secondary (2nd stage) 6.00 –0.78 6.70 –0.14 12.70 –0.91 0.91
Voc Tr. 11.80 8.11 9.10 5.32 20.90 13.43 13.43
Higher 10.00 –3.44 12.00 –2.76 22.00 –6.20 6.20
49.40 50.50 99.90
SD 0.75 –0.85 –0.10
SAD 33.79 27.32 61.10 61.10
1 Percentage of men without formal education with respect to the total interviewed in the European Social Survey.
2 This value was obtained by subtracting the distribution of the universe from the distribution sample. The positive values indi- cate an over-represented sample, and the negative values indicate an under-represented sample with respect to the universe.
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 35
TABLE 2. Comparison between the sample and the universe in the distribution of educational level and sex. Ver- tical percentages and differences between magnitudes (sample minus universe) (Continuation)
Role of Government/Work orientations
Men Women Total
Data Difference Data Difference Data Difference SVA
No formal education 4.60 1.22 5.50 0.39 10.10 2.01 1.60
Primary 7.80 1.12 9.00 1.13 16.80 2.31 2.25
Secondary (1st stage) 11.40 –3.68 11.10 –2.29 22.50 –6.20 5.97
Secondary (2nd stage) 6.00 –0.78 6.70 0.19 12.70 –0.66 0.97
Voc Tr. 7.40 3.79 7.70 3.87 15.10 7.66 7.66
Higher 9.70 –3.46 13.00 –1.58 22.70 –5.12 5.04
46.90 51.20 100.00
SD –1.79 1.72 0.00
SAD 14.04 9.45 23.96 23.50
Health Barometer (2nd wave)
Men Women Total
Data Difference Data Difference Data Difference SVA
No formal education 2.20 –1.18 3.20 –1.91 5.40 –2.79 2.79
Primary 8.60 1.92 10.40 2.53 19.00 4.51 4.51
Secondary (1st stage) 12.40 –2.68 9.70 –3.69 22.10 –6.60 6.60
Secondary (2nd stage) 7.70 0.92 7.60 1.09 15.30 1.94 1.94
Voc Tr. 8.20 4.59 8.30 4.47 16.50 9.06 9.06
Higher 9.50 –3.66 12.00 –2.58 21.50 –6.42 6.42
48.60 51.20 99.80
SD –0.09 –0.08 –0.30
SAD 14.95 16.28 31.32 31.32
Source: See table 1. Data of the universe taken from the INE: 2017b and 2016d.
The comparison of employment variables
Only labour force participation and unemployment rates were taken into ac- count in the question about employment; other responses pertaining to groups that were not related to the aims of this study were not considered. Two of the three surveys used asked respondents about
their employment status, and consid - ered them to be unemployed only when they said that they were; in contrast, the LFS defined employment status and un- employment based on several questions that allowed the “inactive” to be differ- entiated from the (actual) unemployed. In other words, in the surveys under study here, unemployment was an allocated
36 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
category, even though respondents may have actually been inactive, and there- fore not within the labour force.
Table 3 shows that two of the three surveys over-represented the labour force participation rate, whereas the ISSP had the greatest differences, amounting to
four percentage points. The results of the European Social Survey had the best fit, followed by the Health Barometer. All three surveys showed smaller differences in men than in women, which implied that they recorded male labour force participa- tion better.
TABLE 3. Comparison between the sample and the universe in labour force participation rate and unemploy- ment rate. Vertical percentages and differences between magnitudes (sample minus universe)
European Social Survey 2017 (Round 8)
Men Women Total
Data Difference Data Difference Data Difference
LF part.rate 64.10 –0.56 55.90 2.64 58.70 –0.11
Unempl. Rate 10.20 –6.23 15.50 –4.28 13.50 –4.49
Role of Government/Work orientations
Men Women Total
Data Difference Data Difference Data Difference
LF part.rate 67.80 1.89 60.50 6.09 63.90 3.99
Unempl. Rate 22.40 4.89 24.20 2.58 30.40 3.90
Health Barometer (2nd wave)
Men Women Total
Data Difference Data Difference Data Difference
LF part.rate 66.00 0.79 57.30 3.39 61.50 2.09
Unempl. Rate 28.10 9.69 31.00 9.18 29.50 9.50
Source: See Table 1. Data of the universe taken from the INE: 2017c and 2016e.
The unemployment rate was under-repre- sented by the ESS by 4.5 points, which gave a worst fit for men. The other two surveys over-represented unemployment rate, partic- ularly in the case of men, although the Health Barometer showed a total difference that al-
most TRIPLED the one found in the ISSP.
The ISSP provided the best fit for unemploy-
ment rate7.
7 SAD values were not presented because they are meaningless, since they are different variables.
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 37
Regarding the occupational situation shown in Table 4, the greatest differences were found among wage-earners, 85% of the labour force participation in Spain. The ESS represented the situation best; the most important aspect was the over-repre- sentation of wage-earning women by just over two points, and the under-representa- tion of self-employed men by 2.13 points. These two situations represented 73.4% of the total variation.
The differences increased slightly in the ISSP as a result of the higher over-represen- tation of wage-earning women and the un- der-representation of male wage-earners. Although both deviations were compensated for (total deviation of 0.35), they represented a change (SAD) of 7.70 points. In fact, these two situations explained 77.8% of the total variation shown in the table. The Health Ba- rometer showed the same tendency, although it was the one that showed the worst fit.
TABLE 4. Comparison between the sample and the universe in occupational status by sex. Vertical percentages and differences between magnitudes (sample minus universe)
European Social Survey 2017 (Round 8)
Men Women Total
Data Difference Data Difference Data Difference SAD
Wage-earners 43.30 –0.43 41.70 2.35 85.00 1.93 2.78
Employers 3.60 –0.08 1.40 –0.16 5.00 –0.24 0.24
Self-employed 5.20 –2.13 4.70 0.96 9.90 –1.17 3.09
Total 52.10 47.80 99.90
SD -2.63 3.16 0.52
SAD 2.63 3.47 3.33 6.11
Role of Government/Work orientations
Men Women Total
Data Difference Data Difference Data Difference SAD
Wage-earners 39.90 –3.68 43.30 4.02 83.20 0.35 7.70
Employers 3.40 –0.13 1.10 –0.47 4.50 –0.60 0.60
Self-employed 7.10 –0.37 5.20 1.24 12.30 0.87 1.60
Total 50.40 49.60 100.00
SD –4.17 4.79 0.62
SAD 4.17 5.73 1.82 9.90
38 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
concluSIonS
The above analysis of the information collected from the three surveys shows that the differences in the age and sex distribu- tion of the probability samples were mark- edly greater than those provided by route and quota samples, as stated in the first hy- pothesis. The sum of absolute deviations of the sample by quotas (3.52) reached 9.05 in the ISSP sample and was almost quadru- pled in the European Social Survey (13.6).
More notable were the mismatches be- tween both probability samples, which could be attributed to the “extra” resources used by the ESS to increase cooperation. Although both made several visits to a home, the ESS rewarded respondents and used refusal conversion strategies, thus re- covering 12.7% of the sample8. The differ- ences in representing the universe are likely to be explained by the specific character- istics of each group (Riba, Torcal and Mo- rales, 2010).
8 In other words, 12.7% of the respondents were in- cluded after several visits to the home using refusal conversion strategies.
The better representative quality of the Health Barometer compared to probability sampling can be explained by the method of selection of the ultimate respondents, us- ing sex and age quotas. In the light of this information, when interviewers “replaced the quota with the neighbouring one” when it was difficult to locate a certain individ- ual (footnote number 9), these replace- ments were compensated for the changes made—in the opposite direction—by other interviewers, so they can be defined as op- posing random errors.
The education level showed greater de- viations, between 23.5 and 60.1 points, mainly produced by the over-representation of people with a lower educational level; in line with what was detected by research carried out in other contexts (Stoop, 2012). Something similar occurred with those who had Vocational Training qualifications. In contrast, respondents who had second- ary or higher education were under-repre- sented, probably caused more by the diffi- culty in contacting them than by any evident desire not to respond (among others, Pasa- das et al., 2006; Beullens et al., 2018; de Leeuw et al., 2018). Regarding this group, the ISSP sample was the one with the least
TABLE 4. Comparison between the sample and the universe in occupational status by sex. Vertical percentages and differences between magnitudes (sample minus universe) (Continuation)
Health Barometer (2nd wave)
Men Women Total
Data Difference Data Difference Data Difference SAD
Wage-earners 39.20 **–4.38 43.30 **4.02 82.50 –0.35 8.40
Employers 2.50 –1.03 2.20 0.63 4.70 –0.40 1.66
Self-employed 7.00 –0.47 5.90 1.94 12.90 1.47 2.40
Total 48.70 –5.87 51.40 6.59 100.10
SD –5.87 6.59 0.72
SAD 5.87 6.59 2.22 12.46
Source: See table 1. Universe data taken from the INE: 2017c and 2016e.
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 39
differences; the other two showed a simi- lar outcome, which meant that the second hypothesis of the greater presence of more highly educated people in probability sam- ples could not be accepted. The under-rep- resentation of young people in the proba- bility samples (5.7 points in the ESS in the 25 and 44 years-old age group, and of just over two points in the under 34 years-old age group in the ISSP) may partially explain the results.
The situation changed completely when employment-related variables were com- pared, as the ESS showed an excellent fit in labour force participation rate, better in men than in women, and an under-repre- sentation of the unemployment rate by 4.4 points. The ISSP had the worst fit, as it over-represented the labour force participa- tion rate and unemployment rate by 4 and 3.9 points, respectively. The Barometer im- proved the estimation of the labour force participation rate but doubled the unem- ployment rate (9.5 points).
Regarding the occupational situation, the probability surveys (especially the ESS) produced a better fit; the greatest differ- ences were found among wage-earners, contrary to what was proposed in the fourth hypothesis.
The methodological design of the ba- rometer should be considered to account for these differences, as homes in which no one responded were replaced with the neighbouring dwelling (Díaz de Rada, 2015). So, whereas in the probability samples sev- eral calls were made before replacing the selected home, the Barometer increased the probability of selection of the homes in- habited during the interviewer’s visit. Since wage-earners spend less time in their home than those who are unemployed, the prob- ability of knocking on a door where there is no answer was higher among those in employment than among the unemployed population. In our opinion, this explains the
higher unemployment rate found by the Health Barometer (and the rest of similar surveys), in line with the third hypothesis.
These findings provide food for thought, firstly, about the resources used by prob- ability sampling. The data availability from the Population Register provided by the Spanish Institute of Statistics is only rarely available, and is far from normal use in the private sector of market and opinion re- search (Díaz de Rada, 2015). In addition, the length of the fieldwork should be taken into account: more than four months for the European Social Survey and almost three months for the ISSP, compared to the nine days for the Health Barometer. Probability sampling requires more time to make con- tact, which may explain these differences. Some experts (among others, Staveren, 1990) have calculated these differences to be twice as long as our quota sampling sur- veys, a proportion that was exceeded in the surveys analysed here.
Changing the quota method for a within- household random selection might –possi- bly– reduce the large differences in ‘labour rates’ in order to adjust them to the uni- verse, as has been found in studies con- ducted in other contexts (among others, Gaziano, 2005; Marlar et al., 2018). This could be done by limiting revisits to four, especially considering that the analyses carried out in Spain with the European So- cial Survey cast doubt on the real effective- ness of the fifth and subsequent contacts (Torcal et al., 2006). These revisits, even if their number and the overall fieldwork time were limited, would extend the length of time spent on fieldwork, as well as an in- crease in the costs of the survey compared with the current situation. To facilitate the operability of the process, a “limit” of revis- its could be set, for example, to contacts made during the first two weeks. Another possibility (operationally more complex) is for the next survey to try to carry out inter-
40 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
views in the homes that were not located in the previous survey.
Finally, after having analysed the infor- mation on the variables considered, the labour deviations detected in the quota sampling could be addressed in the col- lection by adding a third quota relating to the employment status of the respondent. In this way the advantages found in the classification by sex and age and in edu- cational level could be maintained, with- out altering the costs and the time spent in collection.
In order to verify whether these results have only shown a specific situation or a generalisable phenomenon, another com- parison was made with the seventh wave of the ESS (2014) and Survey No. 3020 by the CIS (ISSP Citizenship), which had the same findings. Díaz de Rada and Martínez (2014) found similar conclusions in a comparison with the fifth wave of the ESS, the CIS 2837 Survey on Environment (ISSP II), and the Health Barometer.
bIblIograPhy
Alvira, Francisco (2011). La encuesta: una perspec- tiva general metodológica. Madrid: Centro de In- vestigaciones Sociológicas.
American Association for Public Opinion Research- AAPOR (2016). Standard Definitions. Final Disposi- tions of Case Codes and Outcome Rates for Sur- veys. Available at: https://www.aapor.org/AAPOR, access September 13, 2018. (9th edition).
Bauer, Johanes J. (2016). “Biases in Random Route Surveys”. Journal of Social Statistics and Meth- odology, 4: 263-287.
Beullens, Koen; Loosveldt, Geer; Vandenplas, Car- oline and Stoop, Iinkele (2018). “Response Rates in the European Social Survey: Increasing, De- creasing, or a Matter of Fieldwork Efforts?”. Survey Methods: Insights from the Field, 1-12. Available at: https://doi.org/10.13094/SMIF- 2018-00003, access December 20, 2018.
Bréchon, Pierre (2015). “Random Sample, Quota Sample”. Bulletin of Sociological Methodology, 126(1): 67-83.
Butcher, B. 1995). “Sampling Methods: an Overview and Review”. Survey Methods Centre Newslet- ter, 15(1): 4-8.
Cea D’Ancona, M.ª Ángeles (2012). Fundamentos y aplicaciones en metodología cuantitativa. Ma- drid: Síntesis.
Centro de Investigaciones Sociológicas (2011). Normas generales para la correcta aplicación de la muestra. (Unpublished document).
Centro de Investigaciones Sociológicas (2016a). In- forme metodológico del estudio 3135: Los ciu- dadanos y el Estado / Orientaciones hacia el tra- bajo. (ISSP, módulos 2015/2016).
Centro de Investigaciones Sociológicas (2016b). Los ciudadanos y el Estado (III) / Orientaciones hacia el trabajo (I) (ISSP). (Estudio n.º 3135).
Centro de Investigaciones Sociológicas (2016c). Barómetro Sanitario (segunda oleada del año 2016). (Estudio n.º 3140).
Centro de Investigaciones Sociológicas (2016d). Fi- cha técnica del Barómetro Sanitario (segunda oleada del año 2016). (Estudio n.º 3140).
Cuxart, Anna and Riba, Clara (2009). “Mejorando a partir de la experiencia de la tercera ola de la ESE en España”. Revista Española de Investiga- ciones Sociológicas, 125: 147-168.
De Leeuw Edith; Hox Joop and Luiten Annemieke (2018). “International Nonresponse Trends across Countries and Years: An analysis of 36 years of La- bour Force Survey data”. Survey Insights: Methods from the Field. Available at: https://surveyinsights. org/?p=10452, access January 10, 2019.
Díaz de Rada, Vidal (2008). “La selección de los en- trevistados últimos en encuestas presenciales”. Revista Española de Investigaciones Sociológicas, 123: 209-247.
Díaz de Rada, Vidal (2015). Manual de campo en la encuesta. Madrid: Centro de Investigaciones Sociológicas.
Díaz de Rada, Vidal and Núñez Villuendas, Adoración (2008). Estudio de las incidencias en la investigación con encuesta. Madrid: Centro de Investigaciones Sociológicas.
Díaz de Rada, Vidal and Martínez Martín, Valentín (2014). “Random Route and Quota Sampling: Do they Offer any Advantage over Probably Sampling Methods?”. Open Journal of Statistics, 4(5): 391-401.
Encuesta Social Europea (2018a). ESS8-2016 Docu- mentation report: the ESS data archive (edition 2.0). Available at: https://www.europeansocialsurvey.
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Vidal Díaz de Rada and Valentín Martínez 41
org/docs/round8/survey/ESS8_data_documenta- tion_report_e02_0.pdf, access November 10, 2018.
Encuesta Social Europea (2018b). Archivo de da- tos de la 8.ª edición. Available at: https://www. europeansocialsurvey.org, access November 10, 2018.
Ernst Stähli, Michèle and Joye, Dominique (2016). “Incentives as a Possible Measure to Increase Re- sponse Rates”. In: Wolf, C.; Joye, D.; Smith, T. W. and Fu, Y.-C. (eds.). The SAGE Handbook of Sur- vey Methodology. London: Sage, pp. 425-440.
Gaziano, Cecile (2005). “Comparative Analysis of Within-household Respondent Selection Tech- niques”. Public Opinion Quarterly, 69(1): 124-157.
Instituto Nacional de Estadística (2016a). Encuesta Continua de Hogares: metodología. Madrid: INE.
Instituto Nacional de Estadística (2016b). Encuesta Continua de Hogares. Madrid: INE.
Instituto Nacional de Estadística (2016c). Revisión del Padrón municipal 2016, explotación a 1 de enero de 2016. Madrid: INE.
Instituto Nacional de Estadística (2016d). Encuesta de Población Activa, segundo trimestre. Pobla- ción de 16 y más años por nivel de formación al- canzado, sexo y grupo de edad. Madrid: INE.
Instituto Nacional de Estadística (2016e). Encuesta de Población Activa, segundo trimestre. Tasas de actividad y paro, por sexo y distintos grupos de edad. Madrid: INE.
Instituto Nacional de Estadística (2017a). Revisión del Padrón municipal 2017, explotación a 1 de enero de 2017. Madrid: INE.
Instituto Nacional de Estadística (2017b). Encuesta de Población Activa, resultados de los dos pri- meros trimestres. Población de 16 y más años por nivel de formación alcanzado, sexo y grupo de edad. Madrid: INE.
Instituto Nacional de Estadística (2017c). Encuesta de Población Activa, resultados de los dos pri- meros trimestres. Tasas de actividad y paro, por sexo y distintos grupos de edad. Madrid: INE.
Instituto Nacional de Estadística (2019). En- cuesta de Población Activa (EPA). Available at: http://www.ine.es/dyngs/IOE/es/fichaProg. htm?cid=1259946010306, access June 23, 2019.
Marlar, Jennifer; Chattopadhyay, Manas; Jones; Jeff; Marken, Stephanie and Kreuter, Marken (2018). “Within-Household Selection and Du- al-Frame Telephone Surveys”. Survey Practice, 11(2).
Marsh, Catherine and Scarbrough, Elinor (1990). “Testing Nine Hypotheses about Quota Sam- pling”. Journal of the Market Research Society, 32(4).
Martínez Martín, Valentín (2004). Diseño de encues- tas de opinión. Madrid: Rama.
Menold, Natalja (2014). “The influence of sampling method and interviewers on sample realization in the European Social Survey”. Survey Me- thodology, 40(1): 105-123.
Menold, Natalja (2018). “The Impact of Payment and Respondents’ Participation on Interviewers’ Accuracy in Face-to-face Surveys”. Field Meth- ods, 30(4): 295-311.
Mercer, Andew W.; Kreuter, Frauke; Keeter, Scott and Stuart, Elizabeth A. (2017). “Theory and Practice in Nonprobability Surveys”. Public Opinion Quarterly, 81(S1): 250-271.
Miller, Peter V. (2017). “Is there a Future for Sur- veys?”. Public Opinion Quarterly. 81(S1): 205- 212.
Murgui, Santiago; Muro, Juan and Uriel, Ezequiel (1992). “Influencia de las sustituciones en la cali- dad de los datos en la encuesta de condiciones de vida y trabajo en España”. Estadística Espa- ñola, 34(129): 137-149.
National Research Council (2013). Nonresponse in Social Science Surveys: a Research Agenda. In: Tourangeau, R. and Plewer, T. (eds.). Washing- ton D.C.: The National Academic Press.
Núñez Villuendas, Adoración (2005). “Incidencias de la entrevista personal en la investigación por encuesta”. Revista Española de Investigaciones Sociológicas, 109: 219-236.
Pasadas del Amo, Sara; Soria Zambrano, Sara Mi- caela and Uribe-Echevarría, Marga (2006). “¿Im- porta el prefijo? Una aproximación a las diferen- cias territoriales en las pautas de respuesta a las encuestas telefónicas”. Metodología de Encues- tas, 8: 3-12.
Riba, Clara; Torcal, Mariano and Morales, Laura (2010). “Estrategias para aumentar la tasa de respuesta y los resultados de la Encuesta Social Europea en España”. Revista Internacional de Sociología, 68(3): 603-635.
Rodríguez Osuna, Jacinto (1991). Métodos de Muestreo. Madrid: Centro de Investigaciones Sociológicas.
Scheaffer, Richard; Mendenhall, William and Ott, Lyman (2007). Elementos de muestreo. Madrid: Thomson-Paraninfo.
42 Household Sampling Designs: Differences and Similarities between Probability Sampling and Route and Quota Sampling
Reis. Rev.Esp.Investig.Sociol. ISSN-L: 0210-5233. N.º 171, July - September 2020, pp. 23-42
Smith, Tom M. (2010). “Surveying across Nations and Cultures”. In: Marsden, P. V. and Wright, J. D. (eds.). Handbook of Survey Research. Bingley: Emerald Group Publishing Limited. (2nd edition).
Staveren, Mark van (1990). “Relative Merits of Quota and Random Sampling”. Current Issues in General Population Sampling Joint Centre for Survey Methods Newsletter, 11(1): 5-7.
Stoop, Ineke (2012). “Unit Non-response Due to Refusal”. In: Gideon, L. (ed.). Handbook of Sur- vey Methodology for the Social Sciences. New York: Springer, pp. 121-147.
Sudman, Seymour (1976). Applied Sampling. New York: Academic Press.
Sudman, Seymour and Edward Blair (1999). “Sam- pling in the twenty-first century”. Journal of the Academy of Marketing Science, 27(2): 269-277.
Torcal, Mariano; Morales, Laura and Riba, Clara (2006). “Supervisión y control de calidad del tra- bajo de campo de la Encuesta Social Europea en España: Evaluación y resultados”. Metodolo- gía de Encuestas, 7(2): 75-97.
Williams, Douglas and Brick, Michael (2018). “Trends in U.S. Face-To-Face Household Survey Nonresponse and Level of Effort”. Journal of Sur- vey Statistics and Methodology, 6(2): 186-211.
Worcester, Robert and Downham, John (1986). Consumer Market Research Handbook. Holland: Elsevier.
RECEPTION: March 3, 2019 REVIEW: May 7, 2019 ACCEPTANCE: July 30, 2019
Copyright of Revista Española de Investigaciones Sociologicas is the property of Centro de Investigaciones Sociologicas and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.