1. Pin down a primary dataset (fallback is the PA crash data)
2. Identify a "first pass" inquiry question
3. Create a cause-effect flow chart which identifies your outcome variables and indepenent variables
• Investigate the type and variation range of each variable of interest
• Do we have enough data reported?
• What would be the "analysis data type" -- meaning, even if original data set type is Text, what would I anticipate the end-variable type becoming after cleaning and transforming? (Use open refine to clean/code)
4. Install Open refine
5. Watch these three videos (1, 2, 3) and start playing with Open Refine.

Thanks to Jill for saving the day...

1. Use the online filter mechanism on the PA crash data home to extract a subset of the 2m rows relevant to your study. To get there, open the crash data home then click the gray box "View Data" in the upper right corner, then an interactive tool is loaded, and you can choose Filter and add conditions by column.
2. Create and export a subset specific to your inquiry question and try loading it in OpenRefine and/or python for basic descriptive stats.
3. Remember to increase your memory allowance on open refine with this tutorial

Exploratory data analysis

Use pandas to create a DataFrame of only your most relevant variables, trimming down rows using .loc[] as needed.

Create histograms using .hist(), and simple plots of your core variables, saving the output to a notebook or raw image files to your project repo.