Winter 2016 R Course Final Paper Description

Due on or before 23 December 2016 at 23:59

In your final paper, you will conduct and interpret analyses in R on a dataset of your choice. The goal of this assignment is to reinforce the main topics in the course, and to give you one document that you can look back on to help you in your future R adventures.

Our last two class periods (14 Dec and 21 Dec) are entirely dedicated to giving you time to work on your papers in class. I encourage you to use this time to work on your papers and get help from me and your fellow R pirates.

Can I work with someone else on the same dataset?

Yes! You are welcome to use the same dataset as another student in the class, and you are welcome to work together on your analyses. However, you must each write your own code and text and turn in your own work! If I suspect that you did not write your paper / code, I reserve the right to ask you to explain it to me personally. If you cannot convince me that your code is your own, you will not pass the class.

How do I get a dataset?

You can use any new dataset you’d like (it can’t be one we’ve already used in class). If you have data for a Bachelor’s or Master’s thesis, you are very welcome to use that. Ideally, your dataset shoudl have at least 100 rows and 5 columns – if you have a dataset you really want to analyze with fewer rows or columns, come talk to me.

If you do not have a dataset already that you’d like to analyze, here are a few places to get one.

1. Published psychology datasets. In recent years, more and more researchers have been posting their raw data online. This is a great development for science, and gives you real world datasets to analyse! Here are a few options

• Harvard dataverse: https://dataverse.harvard.edu/dataverse/harvard. Harvard University publishes thousands of open datasets. You can search for one related to psychology.
• Journal of Judgment and Decision Making: http://journal.sjdm.org/. The Journal of Judgment and Decison Making requires all recent authors to publish their data. If you look at issues after 2014, you should find raw data for most if not all published articles.

2. Machine Learning datasets

• UCI machine learning database: (http://archive.ics.uci.edu/ml/datasets.html). This database has 307 datasets from many fields. If you want to use one of these datasets, I recommend you use those with a Multivariate data type and a Regression default task (http://goo.gl/hm2v4B).

• Kaggle: https://www.kaggle.com/datasets. Kaggle is a data science website with hundreds of datasets. Some of these datasets are very, very large. So I recommend you look for one that is on the smaller side.

• R-Bloggers: https://www.r-bloggers.com/r-packages-for-data-access/?utm_source=feedburner&utm_medium=email&utm_campaign=Feed%3A+RBloggers+%28R+bloggers%29. This page has many links to R packages containing data.

3. Other

• ClinicalTrials.gov: https://clinicaltrials.gov/ct2/home. ClinicalTrials.gov has hundreds of clinical datasets.
• Financial datasets: http://www.r-bloggers.com/financial-data-accessible-from-r-part-iii/. Several financial datasets are available here.

How should I write and format my paper?

Your paper will be a combination of text, R code, and R output. Your paper does not need to be formatted in any particular style (though it would be good practice for you to try to approximate APA style). However, you do need to include your name, date, and a title of your paper at the beginning of the document.

Here are three ways you can write your paper (in order of my recommendation)

1. Markdown. Markdown is the most elegant, and easiest way to write documents. With Markdown, you can easily include all of your R code, output, and writing in one document. We’ll go over how to write APA style papers with R code in Markdown later in the course.

2. LaTeX. We won’t cover LaTeX in this course, but if you already use it, you are welcome to use it for this course.

3. Word. This is the least elegant solution. It will require you to copy and paste all of your code and and results from R to Word. However, if you insist on using it you can.

What should be in my paper?

There should be four sections in your paper: Dataset description, Questions, Analyses, and Conclusion.

Task Checklist

1. Read the data into R from a text file using read.table()

2. Show summary statistics of your dataframe using summary() or str().

3. Display the first few rows of your dataframe with head().

4. Print the names of the columns in the dataframe with names()

5. Recode the values of at least one column using indexing and assignment. For example, in a column of sex data, you could change “female” to 1, and “male” to 0.

6. Calculate at least one standard deviation with sd(), one mean with mean(), and one median with median().

7. Print a table of the frequencies of outcomes of a categorical variable with table().

8. Count the number of outliers in one numerical vector. Define an outlier as any datapoint that is more than 3 standard deviations away from the mean.

9. Create at least one scatterplot with plot(). Include a legend with legend() and a regression line.

10. Create at least one histogram with hist()

11. Create at least one pirateplot with pirateplot()

12. Use aggregate() or dplyr to calculate descriptive statistics across groups of data.

13. Calculate at least one t-test with t.test(). Write your results in APA format.

14. Calculate at least one t-test test on a subset of data by combining subset() with t.test(). For example, you could conduct a t-test comparing results from a performance test between two experimental groups, but only for females. Write your results in APA format.

15. Calculate at least one correlation test with cor.test(). Write your results in APA format.

16. Calculate at least one correlation test on a subset of data by combining subset() with cor.test(). For example, you could conduct a correlation between height and weight but only for males. Write your results in APA format.

17. Calculate at least one chi-square test with chisq.test(). Write your results in APA format.

18. Calculate at least one one-way ANOVA analysis with aov() with 1 independent variable. Calculate post-hoc tests for any significant effects. Write your results in APA format.

19. Calculate at least one multiple-variable ANOVA analysis with aov() with 2 or more independent variables. Calculate post-hoc tests for any significant effects. Write your results in APA format.

20. Calculate at least one regression analysis with lm() or glm(). Write your results in APA format.

21. Create (and use!) at least one new custom function. For example, you could create a function called my.histogram() that creates a histogram with colors that you like. Or you could create a function called find.outliers() that looks for outliers in a vector and then tells you where they are in that vector.

For example, in the following, I will ask questions about chicken weights, run the appopriate R code, display the results, and then interpret them.

Was there an effect of diet on Chicken weights? To start my analysis, I calculated the mean weight of chickens separately for each diet.

# Calculate mean weight for each diet
aggregate(formula = weight ~ Diet, 
          data = ChickWeight,
          FUN = mean)

##   Diet   weight
## 1    1 102.6455
## 2    2 122.6167
## 3    3 142.9500
## 4    4 135.2627

My results showed that chickens on Diet 1 had the smallest mean weight (102.65), and chickens on Diet 3 had the highest mean weight (142.95). To see if this effect was significant, I conducted a one-way ANOVA

# One-way ANOVA of weight on Diet
weight.aov <- aov(formula = weight ~ Diet, data = ChickWeight)

# Print the results
summary(weight.aov)

##              Df  Sum Sq Mean Sq F value   Pr(>F)    
## Diet          3  155863   51954   10.81 6.43e-07 ***
## Residuals   574 2758693    4806                     
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

The one-way ANOVA was significant F(3, 574) = 10.81, p < .01, suggesting that diet did affect weights. To understand the nature of this difference, I conducted a post-hoc test

# Post-hoc test on anova

TukeyHSD(weight.aov)

##   Tukey multiple comparisons of means
##     95% family-wise confidence level
## 
## Fit: aov(formula = weight ~ Diet, data = ChickWeight)
## 
## $Diet
##          diff         lwr      upr     p adj
## 2-1 19.971212  -0.2998092 40.24223 0.0552271
## 3-1 40.304545  20.0335241 60.57557 0.0000025
## 4-1 32.617257  12.2353820 52.99913 0.0002501
## 3-2 20.333333  -2.7268370 43.39350 0.1058474
## 4-2 12.646045 -10.5116315 35.80372 0.4954239
## 4-3 -7.687288 -30.8449649 15.47039 0.8277810

Post-hoc tests revealed a significant difference between diets 1 and 3 (p < .05), and diets 1 and 4 (p < .05). No other comparisons were significant at the .05 level.