meta-analysis_exercise.qmd

---
title: "Meta-Analysis Demo"
author: "Psych 251"
date: "10/27/2020"
output: html
---

# Preliminaries

Packages needed for the class example. 

```{r packages}
library(tidyverse)
library(metafor)
library(here)
```

You'd need to run this next chunk if you wanted to re-download the data from MetaLab, which would also mean installing the `metalabr` package (commented out). 

```{r data, eval=FALSE}
# devtools::install_github("langcog/metalabr")
library(metalabr)
ml_dataset_info <- get_metalab_dataset_info()
d <- get_metalab_data(filter(ml_dataset_info, short_name == "mutex"))
d %>%
  select(
    long_cite, short_cite, expt_num, n, d_calc, d_var_calc,
    mean_age_months, year
  ) %>%
  filter(mean_age_months <= 24) %>% # subset for ease
  write_csv(here("data/mutex.csv"))
```

Load in the pre-cached data. 

```{r dataloading}
d <- read_csv(here("data/mutex.csv"))
```

# Basic descriptives

Always take a look at the data first.

```{r examine}
head(d)
```

The effect sizes are in `d_calc` (for Cohen's $d$, calculated from the data). 

```{r hist}
ggplot(d, aes(x = d_calc)) +
  geom_histogram(binwidth = .25) +
  xlab("Effect Size (d)")
```

Since these are developmental data, we can plot them against age. 

```{r ageplot}
ggplot(d, aes(x = mean_age_months, y = d_calc)) +
  geom_point(aes(size = n), alpha = .5) +
  geom_smooth(method = "lm") +
  xlab("Mean age (months)") +
  ylab("Effect size (d)") +
  geom_hline(lty = 2, yintercept = 0)
```

# Meta-analysis

Random effects meta-analysis - this is the default for `metafor`. The main command for metafor is `rma` - that's like the `lm` or `lmer` of the package. 

```{r ranef}
random_effects_mod <- rma(
  yi = d_calc, vi = d_var_calc,
  slab = short_cite, data = d
)
summary(random_effects_mod)
```


For kicks, try fixed effects. 

```{r fixef}
# look up the documentation for rma and figure out how to do this!
```

# Forest plot

`metafor` also lets you create forest plots. 

```{r forest-ranef}
forest(random_effects_mod)
```

Compare to the forest plot for the fixed effects model.

```{r forest-fixef}

```

# Funnel plot

A funnel plot can be used to diagnose publication bias (though it's not the most sensitive way to do so).

```{r funnel}
funnel(random_effects_mod)
```


# Meta-regression

Meta-regression asks whether study-level covariates (like say year of publication or average age of kids) are related to the effect size. 

```{r metareg-age}
meta_reg_model <- rma(
  yi = d_calc, vi = d_var_calc, mods = ~mean_age_months,
  slab = short_cite, data = d
)
summary(meta_reg_model)
```
Try asking if publication `year` is a significant meta-regressor.

```{r metareg-year}

```