• Collaborative Research Project

• Next Class

• Review

• Static maps with ggmap

• Dynamic results presentation

  • Static website hosting with gh-pages

Purposes: Pose an interesting research question and try to answer it using data analysis and standard academic practices. Effectively communicate your results to a variety of audiences in a variety of formats.

Deadline:

• Presentation: In-class 2 December

• Website/Paper: 16 December 2016

The project can be thought of as a 'dry run' for your thesis with multiple presentation outputs.

Presentation: 10 minutes maximum. Engagingly present your research question and key findings to a general academic audience (fellow students).

Paper: 5,000 words maximum. Standard academic paper, properly cited laying out your research question, literature review, data, methods, and findings.

Website: An engaging website designed to convey your research to a general audience.

Project total: 50% of your final mark.

• 10% presentation

• 10% website

• 30% paper

As always, you should submit one GitHub repository with all of the materials needed to completely reproduce your data gathering, analysis, and presentation documents.

Note: Because you've had two assignments already to work on parts of the project, I expect high quality work.

Find one other group to be a discussant for your presentation.

The discussants will provide a quick (max 2 minute) critique of your presentation–ideas for things you can improve on your paper–pose questions.

I will have normal office hours this week and next week.

Please take advantages of this opportunity to improve your final project.

Be prepared.

• What is the basic R syntax for a regression model?

• What is a model function? What two parts do GLM model functions have?

• How do you find a 95% confidence interval for a parameter point estimate (both mathematically and in R)?

• What are some more effective ways to present results from a logistic regression to both statistical and general audiences?

Earlier we didn't have time to cover mapping with ggmap.

We've already seen how ggmap can be used to find latitude and longitude.

library(ggmap)

places <- c('Bavaria', 'Seoul', '6 Pariser Platz, Berlin')

geocode(places)

##         lon      lat
## 1  11.49789 48.79045
## 2 126.97797 37.56654
## 3  13.37854 52.51701

qmap(location = 'Berlin', zoom = 15)

Example from: Kahle and Wickham (2013)

Use crime data set that comes with ggmap

names(crime)

##  [1] "time"     "date"     "hour"     "premise"  "offense"  "beat"    
##  [7] "block"    "street"   "type"     "suffix"   "number"   "month"   
## [13] "day"      "location" "address"  "lon"      "lat"

# find a reasonable spatial extent
qmap('houston', zoom = 13) # gglocator(2) see in RStudio

# only violent crimes
violent_crimes <- subset(crime,
    offense != "auto theft" & offense != "theft" &
    offense != "burglary")

# order violent crimes
violent_crimes$offense <- factor(violent_crimes$offense,
    levels = c("robbery", "aggravated assault", "rape", "murder"))

# restrict to downtown
violent_crimes <- subset(violent_crimes,
                    -95.39681 <= lon & lon <= -95.34188 &
                    29.73631 <= lat & lat <= 29.78400)

# Set up base map
HoustonMap <- qmap("houston", zoom = 14,  
                   source = "stamen", maptype = "toner")

# Add points
FinalMap <- HoustonMap +
                geom_point(aes(x = lon, y = lat, colour = offense),
                data = violent_crimes) +
                xlab('') + ylab('') +
                theme(axis.ticks = element_blank(),
                      axis.text.x = element_blank(),
                      axis.text.y = element_blank()) +
                guides(size = guide_legend(title = 'Offense'),
                       colour = guide_legend(title = 'Offense'))

print(FinalMap)

When your output documents are in HTML, you can create interactive visualisations.

Potentially–though not always–more engaging and could let users explore data on their own.

Big distinction:

Client Side: Plots are created on the user's (client's) computer. Often JavaScript in the browser. You simply send them static HTML/JavaScript needed for their browser to create the plots.

Server Side: Data manipulations and/or plots (e.g. with Shiny Server) are done on a server in R. Browsers don't come with R built in.

There are lots of free services (e.g. GitHub Pages) for hosting webpages for client side plot rendering.

You usually have to use a paid service for server side data manipulation plotting.

You can use R to (relatively) easily create server side web applications with R.

To do this use Shiny.

We are not going to cover Shiny in the class as it usually requires a paid service to host.

You already know how to create HTML documents with R Markdown.

results='asis' in code chunk head (not needed for some packages).

There is a growing set of tools for interactive plotting, e.g.:

• plotly

• googleVis

• dygraphs

• networkD3

These packages simply create an interface between R and (usually) JavaScript.

Debugging often requires some knowledge of JavaScript and the DOM.

In sum: usually simple, but can be mysteriously difficult without a good knowledge of JavaScript/HTML.

The plotly package allows you to convert (most) ggplot2 plots to JavaScript.

Simply create your ggplot2 object, then pass it to ggplotly.

Using an example from last class:

mort_plot <- ggplot(data = MortalityGDP, aes(x = InfantMortality,
        y = GDPperCapita)) + geom_point()

Then . . .

library(plotly)
ggplotly(mort_plot)

plot_ly(MortalityGDP, x = ~InfantMortality, y = ~GDPperCapita,
        mode = 'markers')

The googleVis package can create Google plots from R.

# Create fake data
fake_compare <- data.frame(
                country = c('2010', '2011', '2012'),
                US = c(10,13,14),
                GB = c(23,12,32))

(Example modified from googleVis Vignettes.)

library(googleVis)
line_plot <- gvisLineChart(fake_compare)
print(line_plot, tag = 'chart')

Note: To show in interactive R use plot instead of print and don't include tag = 'chart'.

library(WDI)
co2 <- WDI(indicator = 'EN.ATM.CO2E.PC', start = 2010, end = 2010)
co2 <- co2[, c('iso2c','EN.ATM.CO2E.PC')]

# Clean
names(co2) <- c('iso2c', 'CO2 Emissions per Capita')
co2[, 2] <- round(log(co2[, 2]), digits = 2)

# Plot
co2_map <- gvisGeoChart(co2, locationvar = 'iso2c',
                      colorvar = 'CO2 Emissions per Capita',
                      options = list(
                          colors = "['#fff7bc', '#d95f0e']"
                          ))

CO2 Emissions (metric tons per capita)

print(co2_map, tag = 'chart')

More examples are available at: http://HertieDataScience.github.io/Examples/

Any HTML file called index.html in a GitHub repository branch called gh-pages will become a hosted website.

The URL will be:

http://GITHUB_USER_NAME.github.io/REPO_NAME

Note: you can use a custom URL if you own one. See https://help.github.com/articles/setting-up-a-custom-domain-with-github-pages/

First create a new branch in your repository called gh-pages:

Then sync your branch with the local version of the repository.

Finally switch to the gh-pages branch.

You can use R Markdown to create the `index.html page.

Simply place a new .Rmd file in the repository called index.Rmd and knit it to HTML. Then sync it.

Your website will now be live.

Every time you push to the gh-pages branch, the website will be updated.

Note branches in git repositories can have totally different files from one another.

Example: networkD3

You can create interactive 'dashboards' for displaying an information overview using the flexdashboard package.

flexdashboard builds on R Markdown.

To set a .Rmd file as a flexdasboard, in the header use:

output: flexdashboard::flex_dashboard

Each element of the dashboard is delimited with the Markdown third level header: ###.

You can create different columns and rows with:

Column
-------------------------------------

Row
-------------------------------------

A minimal code example is available at: https://raw.githubusercontent.com/HertieDataScience/flexdashboard_example/gh-pages/index.Rmd

The output is at: http://hertiedatascience.github.io/flexdashboard_example/

You can host these on Github pages as before.

Then can also be integrated with shiny.

Begin to create a website for your project with RMarkdown and graphics (either static or interactive).

If relevant include:

• A table of key results

• A googleVis map

• A bar or line chart with plotly or other package

• A simulation plot created with Zelig or other tool showing key results from your analysis.

Push to the gh-pages branch.