A.1 Number of Establishments Surveyed, by Strata

TABLE A.1 Number of Establishments Surveyed, by Strata

Country Source Sampling frame Year and mode of data collection

Bangladesh Bangladesh Bureau of Statistics Establishment census, 2013a

Brazil Ministry of Labor Establishment census, RAIS, 2018b

Burkina Faso Business Registry Business Registry in Commerce and Industry Chamber

Ghana Ghana Statistical Service

Economic Census (IBES Phase 1 and Phase 2), 2013 India Central Statistics Office of India Economic Census, 2013 Annual Survey of Industries (ASI), 2017–18c

Kenya Kenya National Bureau of Statistics Establishment census, 2017 Korea, Rep. Statistics Korea Establishment census, 2018 Malawi National Statistical Office of Malawi Establishment census, 2018 Poland Statistics Poland Establishment census, 2020

Senegal National Agency for Statistics and Demography Establishment census, 2016

Vietnam General Statistics Office of Vietnam Establishment census, 2018 2019, face-to-face 2019, face-to-face 2021, telephone

2021, telephone

2020, face-to-face

2020, telephone 2020–21, telephone 2019–20, face-to-face 2021, telephone 2019, face-to-face

2019, face-to-face

Source: Original table based on the Firm-level Adoption of Technology (FAT) survey. a. For Bangladesh, the sampling frame was based on the latest establishment census available complemented with an updated list from the business registry. b. For Brazil, the information came from Relação Anual de Informações Sociais (RAIS), a matched employer-employee database covering all formal firms. Data for Brazil are only for the state of Ceará. c. For India, the sampling frame included firms with 10 or more workers and combines the latest establishment census (2013) for services and the ASI (2017–18) for manufacturing. Data for India are only for the states of Tamil Nadu and Uttar Pradesh.

Standard Industrial Classification of All Economic Activities (ISIC), Rev. 4. More specifically, the sample includes firms from the following ISIC Rev. 4 sectors: agriculture (ISIC 01, from Group A); all manufacturing sectors (Group C); construction (Group F); wholesale and retail trade (Group G); transportation and storage (Group G); accommodation and food service activities (Group I); information and communication (Group J); financial services (ISIC 64) (from Group K, financial and insurance activities); travel agency (ISIC 79, from Group N); health services (ISIC 86, from Group Q); and repair services (ISIC 95, from Group S).

The survey was stratified according to the universe of establishments by sector of activity, firm size, and geographic regions. The sample is representative across these dimensions. For sectors, for all countries, the sample was stratified at least for agriculture (ISIC 01); food processing (ISIC 10); wearing apparel (ISIC 14); retail and wholesale (ISIC 45, 46, and 47); other manufacturing (Group C, excluding food processing and apparel); and other services (including all other firms, excluding retail). This sector structure of the data was used for most of the analysis in this volume. Additional sector stratification that was country specific included: motor vehicles (ISIC 29); leather (ISIC 15); pharmaceuticals (ISIC 21); land transport (ISIC 49); financial services (ISIC 64); and health services (ISIC 86). For the firm size stratification, there are three strata: small firms (5–19 workers); medium firms (20–99 workers); and large firms

(100 or more workers). Table A.2 shows the distribution of the universe of establishments by sector (agriculture, manufacturing, and services) and firm size (small, medium, and large). In the geographic stratification, subnational regions are used. To calculate the optimal distribution of the sample, the team followed a methodology described in World Bank (2022). The sample size for each country was aligned with the degree of stratification of the sample. Table A.3 presents the number of firms surveyed by aggregated sector and by firm size.

Survey Weights

FAT surveys are cross-sectional surveys and rely on probability samples. Before starting the survey in each country, an independent and entirely new sample was randomly selected from the most recent and comprehensive sampling frame available. Therefore, for any FAT survey, the initial weights to be attached to sampled units (which are establishments) are design weights: they are equivalent to unit inclusion probabilities. For any given country, the target population of the FAT survey is the population of establishments that (1) exist at the reference time of the survey; (2) are located within a specific set of regions; (3) operate within a specific set of sectors; and (4) have at least 5 workers.

All FAT surveys adopt a stratified one-stage element sampling design. Establishments are randomly selected with equal probabilities within strata, by sector, region, and firm size groups. Because the sample is not proportionally allocated to the strata, inclusion probabilities differ between strata. The statistical analysis of FAT survey data presented in this volume is performed using the weights to properly account for the selection of sample units with unequal probabilities. FAT weights were adjusted for nonresponse by means of a simple Response Homogeneity Groups (RHG) model (Särndal, Swensson, and Wretman 1992), with groups determined by sampling strata.

Because of the different number of establishments in each country, when computing global statistics for the data, weights were rescaled so that all countries are equally weighted. This means that for results between strata presented in this volume that are not country specific, the weights represent the cross-country average, such that each country has similar weights. Technical details about the weights used in the FAT data are described by Zardetto (forthcoming). In addition, given the significant differences in economic structure, formality, and other economic characteristics of the samples included in the FAT survey, regression tools with controls (e.g., size, sector, and country) are used to adjust some of the statistics shown for the whole sample with different countries and sectors, and to facilitate the comparisons.

Implementation, Quality Control, and Validation

A critical objective of the data collection effort is to obtain robust and comparable measures of the sophistication of technologies used across countries, sectors, firms, and business functions. This requires fully harmonized implementation processes across countries that minimize potential nonresponse, enumerator, and respondent biases.