Final Project

Overview

For the final project, your job will be to use the skills you’ve acquired in this class to ask questions of a real-world dataset (or datasets).

You’ll work independently, but we highly encourage you to come to office hours and lab to get feedback on your ideas.

Grading and Deliverables

Your final deliverable (what you turn in) will be a Jupyter notebook with the following sections (See the table below); each section is worth a certain number of points, adding up to 17; the final project proposal is also worth 1 point.

Section	Points	Description	Example 1	Example 2
Introduction	2	What dataset are you looking at? Where/how was it created? What question(s) will you be asking?	Dataset about which construction people use in the dative alternation: do they use NP (“She gave the man the box”) or PP (“She gave the box to the man”). I will ask which features predict the use of one construction vs. the other.
Data	3	Descriptive statistics about the dataset: number of rows/columns, central tendency (mean/median/mode), variability, any missing values, etc. Should also include details of cleaning or merging datasets, should you need to do that.	The dativeSimplified dataset contains 903 observations with 5 variables; it was created by examining transcriptions of conversational data from Switchboard. No cleaning was required.
Visualizations	4	2-3 graphs showing specific patterns or features you’d like to highlight. Each visualization should be accompanied by a short (1-2 sentences) description of what you think it shows.	Boxplot showing length of the theme argument when recipient is realized as a noun phrase vs. prepositional phrase. Barplot showing proportion of NP realizations depending on animacy of recipient.
Analyses	4	2-3 analyses discussed in class (e.g., linear regression, etc.) to address your question. Each analysis should be accompanied by a short (1-3 sentences) interpretation. Should also include evaluation of your model somehow, e.g., \(R^2\), AIC, etc.	Logistic regression predicting realization (NP vs. PP) from Animacy and Length. Compare AIC of this model to a model omitting each variable in turn.
Limitations and Ethical Issues	3	Discuss any limitations to your approach, as well as potential ethical issues (if relevant).	Variables could be inter-related; also only 4 predictor variables total.
Conclusion	1	Drawing a conclusion about the dataset and the questions you posed.	An NP realization is more likely for longer themes.

Suitable datasets

There are two critical requirements for your dataset:

1. First, it should be about Social science data. If you’re unsure whether a given dataset qualifies, feel free to ask us.

2. Second, it should at least 3 variables (but preferably more). We don’t want you to just analyze a single variable, nor how a single variable relates to one other variable. Ideally, you would consider how multiple variables relate (e.g., in multiple regression).

With that in mind, here are some ideas for suitable datasets. You can use one of these (and ask questions that you find interesting), but you can also feel free to find one of your own; that said, these datasets have been vetted a little more.

Dataset	Social Science Domain	Description	Accessing
World Bank Open Data	Economics / Global Development	Contains time series data for many domains, such as agricultural development, rural poverty, carbon emissions, and much, much more.	Link to Data Bank; can browse by “indicator”; may require merging datasets for more information.
World Happiness Report	Economics / Global Development	Dataset about global happiness scores; might need to be merged with other datasets to ask useful questions.	Kaggle
World Energy Consumption	Contains time series data about consumption of energy and electricity.	Link on Kaggle
SCARFS (Spontaneous, controlled, acts of reference between friends and strangers)	Linguistics/Communication	Data about friends and strangers playing the game Taboo, which clues they gave, and whether a trial was wason.	GitHub Link
California Housing Prices	Economics	Information about the median house value for different districts in California.	Link on Kaggle.
Student alcohol consumption	Public Health	Information about student behavior, including alcohol consumption and more.	Link on Kaggle.
Dative Alternation	Linguistics	Dataset about which construction people use in the dative alternation: do they use NP (“She gave the man the box”) or PP (“She gave the box to the man”).	Online, or ask Professor Trott for .csv file

Project difficulty

Note that we will also take into consideration the difficulty involved in your project, especially in the Data stage (e.g., cleaning data, merging multiple datasets).

Finding and merging multiple datasets is not necessary––i.e., you can get full points without it––but if you successfully clean and merge multiple datasets, it could “offset” lower performance on another section (e.g., Visualization); that is, you can think of it as a kind of unofficial extra credit.

Here are some examples of some things that, while not necessary, would make your project more difficult and thus more impressive:

• Merging multiple datasets that you’ve found to leverage data linkage.

• Comparing multiple statistical models using model selection methods we’ve discussed in class.