Interventions to promote technology adoption in firms: A systematic review

Abstract

Abstract Background The adoption of improved technologies is generally associated with better economic performance and development. Despite its desirable effects, the process of technology adoption can be quite slow and market failures and other frictions may impede adoption. Interventions in market processes may be necessary to promote the adoption of beneficial technologies. This review systematically identifies and summarizes the evidence on the effects of interventions that shape the incentives of firms to adopt new technologies. Following Foster and Rosenzweig, technology is defined as “the relationship between inputs and outputs,” and technology adoption as “the use of new mappings between input and outputs and the corresponding allocations of inputs that exploit the new mappings.” The review focuses on studies that include direct evidence on technology adoption, broadly defined, as an outcome. The term intervention refers broadly to sources of exogenous variation that shape firms' incentives to adopt new technologies, including public policies, interventions carried out by private institutions (such as NGOs), experimental manipulations implemented by academic researchers trying to understand technology adoption, and natural experiments. Objective The objective of this review is to answer the following research questions: 1. To what extent do interventions affect technology adoption in firms? 2. To what extent does technology adoption affect profits, employment, productivity, and yields? 3. Are these effects heterogeneous across sectors, firm size, countries, workers' skill level, or workers' gender? Selection Criteria To be included, papers had to meet the inclusion criteria described in detail in Section 3.1 which is grouped into four categories: (1) Participants, (2) Interventions, (3) Methodology, and (4) Outcomes. Regarding participants, our focus was on firms, and we omitted studies at the country or region level. In terms of interventions, we included studies that analyzed a source of exogenous variation in incentives for firms to adopt new technologies and estimated their effects. Thus, we left out studies that only looked at correlates of technology adoption, without a credible strategy to establish causality, and only included studies that used experimental or quasi‐experimental methods. Regarding outcomes, papers were included only if they estimated effects of interventions (broadly defined) on technology adoption, although we also considered other firm outcomes as secondary outcomes in studies that reported them. Search Methods The first step in selecting the studies to be included in the systematic review was to identify a set of candidate papers. This set included both published and unpublished studies. To look for candidate papers, we implemented an electronic search and, in a subsequent step, a manual search. The electronic search involved running a keyword search on the most commonly used databases for published and unpublished academic studies in the broad topic area. The words and their Boolean combinations were carefully chosen (more details in Section 3.2). The selected papers were initially screened on title and abstract. If papers passed this screen, they were screened on full text. Those studies that met the stated criteria were then selected for analysis. The manual search component involved asking for references from experts and searching references cited by papers selected through the electronic search. These additional papers were screened based on title and abstract and the remaining were screened on full text. If they met the criteria they were added to the list of selected studies. Data Collection and Analysis For the selected studies, the relevant estimates of effects and their associated standard errors (SEs) were entered into an Excel spreadsheet along with other related information such as sample size, variable type, and duration for flow variables. Other information such as authors, year of publication, and country and/or region where the study was implemented was also included in the spreadsheet. Once the data were entered for each of the selected studies, the information on sample size, effect size and SE of the effect size was used to compute the standardized effect size for each study to make the results comparable across studies. For those studies for which relevant data were not reported, we contacted the authors by email and incorporated the information they provided. Forest plots were then generated and within‐study pooled average treatment effects were computed by outcome variable. In addition, an assessment of reporting on potential biases was conducted including (1) reporting on key aspects of selection bias and confounding, (2) reporting on spillovers of interventions to comparison groups, (3) reporting of SEs, and (4) reporting on Hawthorne effects and the collection of retrospective data. Results The electronic and manual searches resulted in 42,462 candidate papers. Of these, 80 studies were ultimately selected for the review after screenings to apply the selection criteria. Relevant data were extracted for analysis from these 80 studies. Overall, 1108 regression coefficients across various interventions and outcomes were included in the analysis, representing a total of 4,762,755 firms. Even though the search methods included both high‐income and developing countries, only 1 of the 80 studies included in the analysis was in a high‐income country, while the remaining 79 were in developing countries. We discuss the results in two parts, looking at firms in manufacturing and services separately from firms (i.e., farms) in agriculture. In each case, we consider both technology adoption and other firm outcomes. Authors' Conclusions Overall, our results suggest that some interventions led to positive impacts on technology adoption among firms across manufacturing, services, and agriculture sectors, but given the wide variation in the time periods, contexts, and study methodologies, the results are hard to generalize. The effects of these interventions on other firm performance measures such as farm yields, firm profits, productivity, and employment were mixed. Policy‐makers must be careful in interpreting these results as a given intervention may not work equally well across contexts and may need to be adjusted to each specific regional context. There is great need for more research on the barriers to technology adoption by firms in developing countries and interventions that may help alleviate these obstacles. One major implication for researchers from our review is that there is a need to carefully measure technology adoption.