Step 3. Synthesize

This is a relatively unassuming step. It precedes the ‘summarize’ step where you really dig in and work on data viz and the prepartion for meta-statistics. Nonetheless, it is a relatively critical step that is includes the following three processes.
a. extract evidence from the filtered list of eligible primary studies
b. tabulate the evidence
c. Data QA/QC and ensure you have what you need to do a meta.

Step 3, synthesize, is akin to collect data within the primary research context. You have designed your experiment, selected the study site and target organism, and planned your sampling effort (and sample sizes). However, you are doing an experiment here but synthesizing other completed experiments with a planned and structure approach. You are in part reviewing every study and in part scraping data from each instance.

Open science product

The data from your efforts should be published as an independent data publication. Depending on data type and discipline, select a data repository that assigns a DOI. You have completed a significant synthesis of research by completing this step - even for a total of 10 studies. The mean, variance, and sample size for every level of each treatment from each study should be included in this dataframe and published. It is not uncommon to exclude include the effect size measure because this is derived data.

Case study

A very short meta-analysis of a limited set of studies, at the time, for cushion plants in the alpine and arthropods is a useful example. The primary objective was to explore whether cushion plants can faciliate arthropods. We used this synthesis as a means to identify whether research in this domain was a viable PhD topic. It was not. We published the data on figshare. In a second instance of a limited number of studies, we explored the frequency of study of positive plant interactions on dunes using meta-analysis and also put the data on figshare. When the data were more comprehensive, more traditional journals were interested. Here is an example of a meta-analysis of global change drivers and invasion in grasslands with the data on figshare. All of these data are appropriate for re-analysis, additional syntheses, and updated searches.

We typically publish biodiversity data with richness and species occurrence data on knb.

To build the basic dataframe, we code each study with a unique numeric ID. This facilitates connecting different dataframes and tracking. We record location, categorical classifications that describe the study, and mean, var, and n as separate vectors for each treatment and control within the experimental factor set. Trials as usually encoded with the 1.1, 1.2 notation. These are the norms from within the ecological synthesis community only.

The best to appreciate synthesis data are to see synthesis data!!

library(tidyverse)
library(DT)
#Cushion plant meta-analysis#### 
cushions.metadata <- read_csv("data/cushion-metadata.csv") #we generally keep a short meta-data table in addition to the detailed textbox and descriptions we provide in the open data repositories when we publish
datatable(cushions.metadata)

cushions.data <- read_csv("data/cushion-data.csv")
datatable(cushions.data) #these data were very diverse and challenging - consequently the open data we shared were the derived data you see here

#View(data) #best way to way to inspect entire dataframe

#Positive interactions on dunes meta-analysis#### 
dune.metadata <- read_csv("data/dune-metadata.csv")
datatable(dune.metadata)

dunes.data <- read_csv("data/dune-growthdata.csv")
datatable(dunes.data)

#Grassland and global change meta-analysis####
grassland.metadata <- read_csv("data/grassland-metadata.csv")
datatable(grassland.metadata)

grassland.data <- read_csv("data/grassland-data.csv")
datatable(grassland.data)

#Human health meta-analysis comparing Nedocromil sodium with placebo for preventing exercise-induced bronchoconstriction####
bronchoconstriction <- read_csv("data/bronchocontriction.csv")
datatable(bronchoconstriction) #excellent example of an important meta within evidence-based medicine, note sample sizes and simplicity

Exercise

Build a dataframe for your set of studies from the search and sort meta-steps. Use an spreadsheet app but save as a .csv. Extract mean, n, and variance estimates from a total of 7-10 studies. Note whether effect size measures were reported within each study. Record location of study in lat and long.

Conclusions

To be able to use the formal synthesis tool of a systematic review, topological data from each study are needed including location, method, number of species, hypothesis tested, and relative certainty that the primary study is relevant.
To be able to do a meta-analysis, either a reported effect size from within a study or an estimate of the mean treatment versus control effect, variance, and sample sizes are needed. Regression or correlation coefficients can be used as an effect size measure directly but counts including number of species cannot be used in meta-statistics.
Extract data, build a dataframe, and check data are the three key process for the synthesis step of an open-science workflow for meta-analysis.
The value of an open data set published in a repository cannot be overstated.

Additional resources

The Cochrane Collaboration Handbook is a superb resource for the workflow.
Chapters 3-5 in this handbook for ecology and evolution is also useful.