exp3_analysis_ratings.Rmd

---
title: 'Experiment 3: Ratings Analysis'
date: "`r Sys.Date()`"
output: 
  github_document:
    toc: true
    toc_depth: 3
  pdf_document:
    toc: true
    toc_depth: 3
editor_options: 
  markdown: 
    wrap: 72
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
library(tidyverse)
options(dplyr.summarise.inform = FALSE)
library(magrittr)
library(lme4)
library(lmerTest)
library(buildmer)
```

# Setup

Variable names:

-   Experiment: exp3\_
-   Data (\_d\_)
    -   d = main df
-   Models (\_m\_)
    -   like = Likeability ratings
    -   acc = Accomplishment ratings
    -   imp = Importance ratings
    -   ratings = build model with all 3 ratings

Load data and select columns used in model. See data/exp3_data_about.txt
for more details.

```{r load-data}
exp3_d <- read.csv("../data/exp3_data.csv", stringsAsFactors = TRUE) %>%
  rename("Participant" = "SubjID", "Item" = "Name") %>%
  select(
    Participant, Condition, GenderRating, Item,
    He, She, Other,
    Likeable, Accomplished, Important
  )
str(exp3_d)
```

Center gender rating for names: Original scale from 1 to 7, with 1 as
most masculine and 7 as most feminine. Mean-centered with higher still
as more feminine.

```{r center-gender-rating}
exp3_d %<>% mutate(GenderRatingCentered = scale(GenderRating, scale = FALSE))
```

Set contrasts for name conditions, now weighted to account for uneven
sample sizes. This uses Scott Fraundorf's function for weighted
contrasts. (The psycholing package version doesn't support doing 2v1
comparisons, only 1v1.) Condition1 is Last vs First+Full. Condition2 is
First vs Full.

```{r contrast-coding}
source("centerfactor.R")
contrasts(exp3_d$Condition) <- centerfactor(
  exp3_d$Condition, c("last", "first")
)
contrasts(exp3_d$Condition)
```

Flip ratings from 1=most likeable/accomplished/important to 7=most
L/A/I, to make interpreting models easier, then mean-center.

```{r flip-ratings}
exp3_d %<>% mutate(
  LikeableFlip = recode(Likeable,
    "1" = 7, "2" = 6, "3" = 5, "4" = 4, "5" = 3, "6" = 2, "7" = 1
  ),
  AccomplishedFlip = recode(Accomplished,
    "1" = 7, "2" = 6, "3" = 5, "4" = 4, "5" = 3, "6" = 2, "7" = 1
  ),
  ImportantFlip = recode(Important,
    "1" = 7, "2" = 6, "3" = 5, "4" = 4, "5" = 3, "6" = 2, "7" = 1
  ),
  LikeableCentered = scale(LikeableFlip, scale = FALSE),
  AccomplishedCentered = scale(AccomplishedFlip, scale = FALSE),
  ImportantCentered = scale(ImportantFlip, scale = FALSE)
)
str(exp3_d)
```

# Likeability

Summary statistics:

```{r means-likeable}
summary(exp3_d$Likeable)
summary(exp3_d$LikeableFlip)
sd(exp3_d$Likeable)
```

Does the Likeability rating of the character predict the likelihood of
*she* as opposed to *he* and *other* responses? The maximal model
includes all interactions, then random intercepts by item but not by
participant.

```{r model-likeable, cache=TRUE}
exp3_m_like <- buildmer(
  formula = She ~ Condition * GenderRatingCentered *
    LikeableCentered + (1 | Participant) + (1 | Item),
  data = exp3_d, family = binomial,
  buildmerControl(direction = "order", quiet = TRUE)
)

summary(exp3_m_like)
```

-   Characters who are rated as more Likeable are more likely to be
    referred to with *she*

# Accomplishment

Summary statistics:

```{r means-accomplishment}
summary(exp3_d$Accomplished)
summary(exp3_d$AccomplishedFlip)
sd(exp3_d$Accomplished)
```

Does the Accomplishment rating of the character predict the likelihood
of *she* as opposed to *he* and *other* responses? The maximal model
includes all interactions, then random intercepts by item but not by
participant.

```{r model-accomplishment, cache=TRUE}
exp3_m_acc <- buildmer(
  formula = She ~ Condition * GenderRatingCentered *
    AccomplishedCentered + (1 | Participant) + (1 | Item),
  data = exp3_d, family = binomial,
  buildmerControl(direction = "order", quiet = TRUE)
)

summary(exp3_m_acc)
```

-   Characters who are rated as more Accomplished are more likely to be
    referred to with *she*, but this is n.s. after correction for
    multiple comparisons.

-   Interaction between Accomplishment, Name Gender Rating, and
    Condition (L vs F+F), but this is n.s. after correction for multiple
    comparisons, so I'm not going to dig into it.

# Importance

Summary statistics:

```{r means-importance}
summary(exp3_d$Important)
summary(exp3_d$ImportantFlip)
sd(exp3_d$Important)
```

Does the Importance rating of the character predict the likelihood of
*she* as opposed to *he* and *other* responses The maximal model
includes all interactions, then random intercepts by item but not by
participant.

```{r model-importance, cache=TRUE}
exp3_m_imp <- buildmer(
  formula = She ~ Condition * GenderRatingCentered *
    ImportantCentered + (1 | Participant) + (1 | Item),
  data = exp3_d, family = binomial,
  buildmerControl(direction = "order", quiet = TRUE)
)

summary(exp3_m_imp)
```

-   Interaction between Important, Name Gender Rating, and Condition (L
    vs F+F), but this is way too small to be significant after
    correction for multiple comparisons

# All Ratings

```{r model-ratings-all, cache=TRUE}
exp3_m_ratings <- buildmer(
  formula = She ~ Condition * GenderRatingCentered * LikeableCentered *
    AccomplishedCentered * ImportantCentered +
    (1 | Participant) + (1 | Item),
  data = exp3_d,
  family = binomial,
  buildmerControl(direction = c("order", "backward"), quiet = TRUE)
)

summary(exp3_m_ratings)
```