p_hackathon.Rmd

---
title: "P-hackathon"
output: html_document
---

[P-hacking](https://www.ncbi.nlm.nih.gov/pubmed/22006061) and [the garden of forking paths](http://www.stat.columbia.edu/~gelman/research/unpublished/p_hacking.pdf) are terms that were invented to describe ways in which data analysts can obtain a statistically significant result by changing the way the data are processed. Some examples of potential processing steps that might not be questioned are:


* Normalizing data or not
* Removing outliers or not
* Creating factor variables or not
* Adjusting for different sets of covariates
* Restricting the range of values being considered
* Data transformations
* Model choices

To illustrate these concepts we are going to do an exercise that I'm calling a "p-hackathon". We are going to do the p-hackathon with this data set: 

```{r}
library(NHANES)
data(NHANES)
```

Before you start analyzing install the matahari R package

```{r}
devtools::install_github("jhudsl/matahari")
```

Then run this command to start recording everything you do in R. 

```{r}
dance_start()
```

Our study is concerned with the association between income (variable name `HHIncomeMid` and gender `Gender`). A simple linear association is: 

```{r}
tidy(lm(HHIncomeMid ~ Gender ,data=NHANES))
```

The rules of the p-hackathon are: 

(1) The goal is to get the tiniest p-value for association possible between the two variables. 
(2) You can only use any transformation/data change you can justify with plausible statistical reasons. (write them down as you go).
(3) You must keep track of _everything_ you did

When you have finished hacking and have your final p-value save your model as: 

```{r}
lm_final = "Your final model goes here" 
tidy(lm_final)
dance_stop()
matahari_history = dance_tbl()
save(matahari_history,file="lastname_phacking.rda")
```
And post your completed rda to the class slack channel.