LOS MEDIOS Y MERCADOS DE LATINOAMERICA:
DESCRIPTION OF THE SURVEY SAMPLE

In these web pages, we have presented many of the results from the Los Medios y Mercados de Latinoamérica study that is conducted annually by us at Audits & Surveys Worldwide. We have received inquiries about our survey sampling methodology from people who wanted to make sure that they are interpreting the results properly. Therefore, we will present a technical description of the sampling methodology in this page.

Target Universe

The target universe of our study is all persons between the ages of 12 and 64 in 18 Latin American countries (Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay and Venezuela) and the Commonwealth of Puerto Rico. These countries are either Spanish-speaking or Portuguese-speaking. In 1996, it is estimated that there are 315 million persons between the ages of 12 and 64 within this target universe.

Survey Universe

Whilst we would like to survey the entire target universe, there are some practical considerations that forced us to exclude certain portions. These reasons include:

• Inaccessibility: Certain places are too difficult to reach while being sparsely populated at the same time. Examples are the Amazonian regions in several countries, the Andean mountains, the Chaco area in Paraguay, the Atacama desert in Chile, etc.
• Security: Certain areas are under the threat of narco-related and guerilla-led activities, such as in Peru and Colombia.
• Lack of sampling materials: Certain places (such as rural areas in El Salvador and Guatemala) have neither accurate population counts nor area maps, so that it is difficult to draw samples properly.

After excluding these areas from the target universe, our survey universe covers 280 million persons between the ages of 12 and 64, which is 89% of the target universe.

We do not claim that the unmeasured population is similar to the study universe. We expect that the unmeasured population to be rural and poorer, and also less likely to have access to media and consumer products. But that is not to say all the excluded areas are like that. For example, all types of consumer products are available in the towns along the Marginal Forest Highway in the Huallaga Valley of Peru as a result of the cocaine-driven economy. But we would be concerned about the physical safety of our interviewers.

Sample Size

The total sample size for this study in 1996 is about 6600 respondents. This number was arrived at based upon the estimated survey costs that can be covered by the potential sales revenue. The respondents are selected through a multi-stage stratified probability sample, as described in the following.

Stage 1: Sample Allocation by Country

The sample was allocated by country based upon anticipated research needs.

• Three countries (Argentina, Brazil and Mexico) were considered to be the most important, because of their large population bases, high spending power and well-developed media environments. Argentina and Mexico were allocated sample size of 800 each, and Brazil was allocated a sample size of 1700. These sample sizes permit each of these countries to be reported reliably..
• Four other countries (Chile, Colombia, Puerto Rico and Venezuela) were considered to be somewhat important and were allocated sample sizes of 600 each (450 for Puerto Rico), so that they too can be reported separately.
• The remaining countries in South America (Bolivia, Ecuador, Paraguay, Peru and Uruguay) were allocated sample sizes of 100 each (150 in Peru), and they are reported together as 'Balance of South America'.
• The countries of Central America and the Caribbean region (Costa Rica, Dominican Republic, El Salvador, Guatemala, Honduras, Nicaragua and Panama) were allocated sample sizes of 80 (100 in Costa Rica), and they are reported together as 'Balance of Central America and Caribbean'.

Stage 2: Sample Allocation by Geographical Region within Country

Generally speaking, we expect the major metropolitan areas to be more affluent and media-rich. Therefore, in each country, the major metropolitan areas are oversampled as separate strata, and the balance of the country forms another stratum. For example, Buenos Aires accounts for 36% of the population of Argentina but we allocated 50% of our sample there. Here are our major metropolitan areas by country:

• Argentina: Capital Federal and Greater Buenos Aires, Cordoba, Mendoza, Rosario
• Bolivia: La Paz, Santa Cruz
• Brazil: Belo Horizonte, Brasilia, Curitiba, Fortaleza, Porto Alegre, Recife, Rio de Janeiro, Salvador, São Paulo
• Chile: Gran Santiago Concepción, Viña del Mar/Valparaíso
• Colombia: Santa Fe de Bogota, Barranquilla, Cali, Medellin
• Costa Rica: San Jose
• Dominican Republic: Santo Domingo, Santiago
• Ecuador: Guayaquil, Quito
• El Salvador: San Salvador, Santa Ana
• Guatemala: Guatemala City
• Honduras: Tegucigalpa, San Pedro Sula
• Mexico: Mexico City, Guadalajara, Monterrey
• Nicaragua: Managua
• Panama: Panama City
• Paraguay: Asunción
• Peru: Lima, Arequipa, Trujillo
• Puerto Rico: Greater San Juan, Mayaguez, Ponce
• Uruguay: Montevideo
• Venezuela: Federal District and Caracas, Barquisimeto, Maracay, Maracaibo, Valencia

Stage 3: Sample Allocation by Socio-Economic Level within Geographical Region

Each geographical region is divided into political/administrative areas. These may be known under various kinds of names (provinces, cities, towns, municipalities, villages, school districts, electoral districts, etc.). Whatever they are called, they may be classified into socio-economic levels, which may be available through government sources, census statistics or other commercial sources.

There is no single classification scheme for socio-economic level that is used in all Latin American countries. There are differences in the number of socio-economic levels, the size of the groups, and the ingredients of the definitions too. In each country, we use the prevailing system there. Ultimately, our objective is to obtain four strata with the following sample allocations:

• 30% in middle-upper and upper class areas
• 40% in middle-middle class
• 20% in middle-lower class
• 10% in lower class

For each of these socio-economic level, we arrange all the political/administrative districts in order of population size. From this list, we draw a random systematic sample of sample locations (or field clusters). Depending on the country, this may occur in a single stage, or in multiple stages (e.g. select the provinces first, then select the cities within, then select the electoral districts within, then select the street blocks within, etc).

Stage 4: Sample Selection of Field Clusters

A field cluster is a sample location at which a group of interviews are conducted. Generally speaking, this corresponds to a street block. At each cluster, a total of 10 interviews, five with women and five with men, are completed. As it should be obvious, the number of clusters is the total number of interviews divided by 10.

The interviewer is provided with a street map of the field location. A random starting point (e.g. third house west starting from the southeast corner) is assigned to each block to indicated where the interviewer should start contacting households. In addition, the interviewer is given a specific path (as indicated by directional arrows drawn on the map) to follow in the designated area in order to complete the assigned number of interviews.

Stage 5: Respondent Selection

Only one household member will be interviewed.

• When there is no one between the ages of 12 and 64 in a household, no interview takes place.
• When there is only one person between the age of 12 and 64, then that person will be interviewed. If the person is not available at the time of initial contact, then an appointment will be scheduled. Two callbacks will be attempted.
• When there are two or more people between the ages of 12 and 64 in the household, then one of them will be randomly selected using the Kish grid method. If the designated person is not available at the time of initial contact, we do not just conveniently interview whoever is present. Rather, an appointment with the designated person will be scheduled, with two callbacks being attempted.

A field supervisor will validate the fieldwork (between 10% to 100%) by retracing the route, verifying household composition and respondent selection as well as checking key survey answers.

Key Features

Our survey sample is based upon the classical probability sampling approach which is the accepted paradigm in most parts of the world. The key feature is that the persons in the study universe have known (but not necessarily equal) probabilities of being selected in the sample. During the processing of the survey results, we make sure that the unequal probabilities of selection are properly accounted for through appropriate weighting. This is what permits us to project our sample results back to the study universe.

It is important to point out what this study is NOT:---

• This is not a quota sample in which the interviewers have discretion as to whom they can choose to interview as long as they deliver the required quota of interviews. The selection procedures are rigorously defined and subject to quality control checks.
• This is not a survey of cable or magazine subscriber study based upon rosters supplied by operators or publishers. This is a general population study.
• This is not an intercept study done at street corners or shopping malls, where the quality and the representativeness of the sample is unknown.
• This is not a telephone interview of an elite sample in countries where telephone access may be limited.
• This is not a mail survey in countries where mailing lists are incomplete and inaccurate, and postal systems are inefficient.
• This is not a survey of self-selected respondents, such as the visitors to a World Wide Web site or some such. The respondents are selected by a precise, objective procedure.
• This survey process can be audited independently after the fact as to whether or not we do what we say we do. 

Validation

Los Medios y Mercados de Latinoamérica is presently the only pan-regional study of its kind in Latin America. There are some studies that claim to be pan-regional in nature, but are in fact based upon the major cities in a few countries. In a previous paper (Major Urban Area Studies), we have shown how such results can be quite misleading. So there is no independent validation of our pan-regional results.

Even within specific countries, our study tends to cover a broader population than the standard national studies. Perhaps the most carefully conducted research studies are the large enumeration surveys done by the television ratings companies in various Latin American countries. Necessarily, we have smaller sample sizes when we match our sample survey against their measured universes (e.g. Buenos Aires/Rosario in Argentina; Mexico City in Mexico; etc). Yet, our numbers hold up very well with theirs, considering the fact that these surveys were conducted by different research companies using different methodologies and procedures. In fact, for one cable system, our estimates were extremely close to each other but significantly lower than what the cable system itself claims, a fact that led to the revelation that the cable system was in fact offering the theoretical number of subscribers, including those who are delinquent on their payments and temporarily cut off.

(posted by Roland Soong on 10/13/97)

(Return to Zona Latina's Home Page)