The dataset includes information about different types of climate change adaptation interventions and their effects on different types of outcomes in the agricultural and coastal sectors in developing countries.

METHODOLOGICAL INFORMATION

Description of methods used for collection-generation of data: This systematic review selected studies that were already included in an evidence gap map (EGM) on adaptation, which is one of the most up-to-date and comprehensive databases on the effectiveness of adaptation interventions in low and middle-income countries (LMICs) (Doswald et al., 2019). The EGM followed the systematic map protocol, which followed guidelines set out by the Centre for Evidence-Based Conservation,(Pullin et al., 2018) and included quantitative or mixed-methods studies and systematic reviews in the analysis. The inclusion criteria for this meta-analysis were adapted from previous research(Doswald et al., 2019) following the PICOS standard (Eriksen & Frandsen, 2018) In a previous study (Doswald et al., 2019) we systematically searched databases of peer-reviewed literature (Web of Science, Scopus, 3ie database and CEE library) and grey literature from several organisational websites for studies on climate change adaptation in low- and middle-income countries (LMICs) as defined by the Organisation for Economic Co-operation and Development (OECD). All literature that had an English abstract and was written in English, Spanish, French or German was included. This yielded a sample of 13,121 studies. The sample was narrowed down by excluding books, book sections and conference proceedings and screening abstract and tiles following several exclusion criteria. Importantly for our purpose, all studies that did not report on the effectiveness of an adaptation intervention were excluded. This yielded a final set of 463 studies (Figure 4), which is published as an interactive EGM at the website of the International Initiative for Impact Evaluation (3ie) (Noltze et al., 2023). In the previous study (Doswald et al., 2019) we categorized studies into four sectors of i) Water, ii) Forestry, fishing and agriculture, iii) Land-use and built environment, and iv) Society, economy and health. Since we focused only on the coastal and agricultural sectors, we excluded 152 studies from their database that did not match these two sectors. We focused on the agricultural and coastal sectors for several reasons. The agricultural sector, along with the forest sector are most directly related with development in LMICs due to the importance of rural areas and the primary sector for those countries’ economies. The forestry sector is critical for climate change mitigation, but the impact of climate change on forest activities has been less documented than in the agricultural sector. The coastal sector has been pioneering in studies of climate change and additionally allowed us to capture intervention effects in urbanized areas. Also, interventions in the coastal sector have tended to target risk reduction outcomes, so by including those interventions we are able to widen the diversity of outcomes studied (the agricultural sector tends to include development-related outcomes). We also excluded primary or non-review studies on NbS or technological interventions in the agricultural sector, and studies which did not have sufficient data for coding. Primary studies on NbS or technological interventions in the agricultural sector were excluded due to the disproportionately large number of systematic reviews for these interventions, which we included in the review. This led to a final batch of 103 included studies, 19 and 84 of which belonged to the coastal and agricultural sectors, respectively. To code the data from the articles, we employed a rigorous qualitative consensus approach(Cascio et al., 2019) to ensure the reliability of our coding. This involved clear coding guidelines, regular communication among coders and iterative discussions to reach agreement. The coding included two stages. First, all three coders coded 6 studies collaboratively until agreement reached saturation; all coders coded the same study and discussed their codes, one study after the other, until coders reached a similar understanding of the variables (i.e., until coders had the same codes of the intervention type, outcome, and effects direction and size variables for two studies in a row). Then, the database of studies was split among the coders and each of them coded her/his batch independently. Questions at this stage were nevertheless solved collaboratively. This strategy enabled us to maintain a high level of coding consistency, enhancing the validity of our study's findings. The database had a hierarchical design: one study could include multiple observations, which were the combination of one intervention and one effect of this intervention in an outcome. Table 2 includes the final number of studies and observations per sector. To measure effects, we looked at the direction and size of effects (“Effects direction” and “Effect size” variables), and the statistical significance of the findings. Direction was coded as positive, neutral or negative. Neutral was coded when the effects were not significant, or the author explicitly mentioned that there were no effects. Effect size was coded via an ordinal scale (“small”, “medium”, “high”) whenever the effects direction was positive or negative (Creutzig et al., 2022) Coding effect sizes required translating the quantitative measures such as means, non-parametric tests, regression coefficients into our ordinal scale. Whenever author´s complemented quantitative metrics with qualitative comments about the size we used the latter. In the studies where authors did not qualify effects as being “small”, “medium” or “high” we assumed that the size of the effects was “medium". The only exception to this rule were observations where the metric value was very small (this was the case for <0.1 beta regression coefficients, mean differences, average treatment effects, and <5% percentage differences between the intervention and control groups). Out of the 103 studies of the database, 11 contained more than 5 observations per study (marked with an asterisk in Supplementary Table 1) and 9 of them contained more than 3 observations about one intervention and outcome type indicating the same effects direction (marked with two asterisks in Supplementary Table 1). Although many observations from single studies may be less generalizable, our unit of analysis was the case and not the study and we weighted all observations equally.