Term Project + Homework 02

• Proposal due on Friday March 20th.
• During exam week: 20 minute presentation of your work. There will be 4 sessions where in groups of 5 you will present to each other and me.
• Write-up will be due on Monday of that week.

• Guiding principle: Use the materials learned in this class. Ex: a two-sample t-test is not enough:
  • Data visualization
  • Data manipulation (not for its own sake)
  • Topics to be covered for sure later: time series, spatial statistics, text mining, and machine learning

• Grading rubric for homeworks applies
• Reproducible research: start with rawest possible data file, set code up so that any one can reproduce the entire analysis with one click of the mouse.

• Has to involve REAL data
• Pick a topic you're interested in (politics, sports, science, etc) and find an interesting question. The more interested you are in an answer, the less it will seem like work!

• Seniors: you can use thesis data. However I require that
  • Your advisors email me and consent to you using your data for both your thesis and this class.
  • Your data processing and analysis be done in R.
• If you're looking for ideas, come talk to me.

Question 4 on page 76 from Chapter 4 of Data Analysis Using Regression and Multilevel/Hierarchical Models. The codebook can be found here.

Using the OkCupid data, fit what you think is a good predictive model for gender and interpret all results.

What do I mean by “good predictions”? Recall that the dataset from class is a sample of profiles. In fact, I gave you 10% of all profiles. I define “bad” as the following: the proportion of people misclassified (both possible ways) using the other 90% of profiles.

Think about what happens when your prediction mechanism is overly optimized for your particular dataset. What will happen when it tries to predict other datasets?

Data consists cancer counts of the National Cancer Institute's “Surveillance, Epidemiology, and End Results Program” (SEER) database of cancers

• For the \( n=887 \) census tracts in the 13 counties in Western Washington: Clallam, Grays Harbor, Island, Jefferson, King, Kitsap, Mason, Pierce, San Juan, Skagit, Snohomish, Thurston, and Whatcom.
• Stratified by different demographic categories: age, sex, and race
• For the years 1996 through 2005

Unprocessed data file as provided by SEER (please do not share). Marked as txt file, but actually is CSV file:

library(dplyr)
SEER <- read.csv("Space Time Surveillance Counts 11_05_09.txt", header=TRUE) %>% tbl_df()
SEER

Let

• \( y_i \) for \( i=1, \ldots, n \) be counts of a particular cancer summed over all years (1996-2005) and over all demographic categories.
• Predictor variables \( X_i \) be relevant demographic factors like age, race, gender, plus socio-economic status and any other relevant variables from the 2000 census.

Keep a tone of where you would be sharing your findings with state public health officials:

• Perform exploratory data analyses: basic summaries, relevant maps (more on mapping making on Monday)
• Conduct appropriate regressions along with interpretations

Do this for breast and lung cancer separately.

Go to Social Explorer while at Reed or using a proxy.

• Click on “Start Now” -> 3-Bar button on top right
• Zoom into WA by entering “Washington” in top left bar
• Select
  • “Census Tract” level of refinement
  • 1. Census 2000 survey
  • 2. Any particular topic
  • 3. Census tracts using “Circle” tool
• Click “Create Report” -> “Data Download” and download “Census Tract data (CSV)” & “Data dictionary (text file)”