2.2. Eligibility criteria

The selection process was carried out using the R package revtools [45] in R software [46]. 5262 studies published between 1965 and September 2022 were identified from the databases (Fig. 1). The first step was to filter duplicated studies, resulting in the exclusion of 1445 articles. To keep relevant publications, the second step involved a title screening, wherein titles meeting the search keywords were selected, leading to the exclusion of 3090 studies. The third step was to filter by abstract, manuscripts had to meet at least one eligibility criteria, i) hairy vetch biomass production at the end of the season (absolute values), ii) maize grain yield (absolute values), and iii) report a treatment with hairy vetch and a control treatment without hairy vetch in the experimental design. A total of 607 papers were removed by this step. Finally, publications were manually selected based on their full text, retaining only the maize studies (excluding 78 articles) and those meeting all the eligibility criteria (excluding a further 19 articles). In October 2022, 23 papers were retained in this last screening.

Flow diagram describing the number of papers collected and various stages of filtering and selection.

The study selection method involved the review of titles and abstracts by one review author, MPR. If necessary, inconsistencies were discussed until a consensus was reached with the other authors, including JV, AJPC, AAC, and IAC. Two review authors, MPR and JV, then independently selected full-text articles for inclusion. In case of disagreement, a consensus on inclusion or exclusion was reached through discussion, and if necessary, a third author, IAC, was consulted.

A step of data filtering, and quality check was executed. A model was fitted considering all observations (n = 147) of 23 selected articles. First, a sensitivity analysis carried out using the leave1out function from the R package metafor, showed that a few observations introduced undesirable residual heterogeneity into this model [47] (Fig. S1). Secondly, an inspection of Cook distances and hat values revealed that those similar observations presented high values, outliers, and leverage points. Lastly, when revising the original source (paper), those observations within each publication reported crop failure, confounding the effect of the cover crop on maize yields were deleted (one observation from Ref. [48] and one observation from Ref. [49]). Other studies [50,51] showed a residual heterogeneity but were not excluded as they did not report in the published manuscript any experimental problems. Finally, the total count remained at 23, as there was no need to completely exclude any publication, and the total number of observations was 145 (Fig. 1). All selected experiments were performed under field conditions. Table 1 describes the main characteristics of these studies.

Description of the publications selected for the analysis. The ID refers to an identification number, citation include the first author and year of publication, year of study refers to the range of years when the experiment took place.