Description

Kitu Komya
Final Report
by Kitu Komya
Introduction
Data
Describe the individual data files
The edmunds original data dimensions is 2445924 by 24. After cleanup, it was 2445924 by 16. The irs original data dimensions is 288 by 111. After cleanup, it was 288 by 7. The cell original data dimensions is 9248 by 22. After cleanup, it was 9248 by 22.
The irs data set is aggregated tax information from the IRS directly from the year 2014. They are based on returns from residents from different zipcodes. The variables here are all numeric since they relate to the number of tax returns.
The cell data set contains information about cell towers in Los Angeles. The data set includes information on the cell tower’s location and type, and thus the variables range from descriptive to numerical.
Clean the data

Summarize

make most_popular_makes
honda 440720
toyota 353630
ford 144697
nissan 118091
subaru 106417

make least_popular_makes
isuzu 1
Oldsmobile 1
geo 3
make least_popular_makes
Isuzu 3
Lotus 3
oldsmobile 3
As we can see from the outputted table, the most popular make is honda, and the least popular make is isuzu and oldsmobile.

What an ugly graph. However, I will not change the x value range because that is omitting data. As we can see, the distribution of suggested price for car to be sold by car manufacturer is highly concentrated at the end range at about 30000. The mean is also at 30000. Although the spread ranges from 0 to 30000, the values before 30000 are basically irrelevant in comparison to the 30000 values. This makes sense because the price suggested by the car manufacturer for a vehicle to be sold is pretty high and around those values.
In this IRS data, we discover a few relationships between the variables. Firstly, the N1 refers to the number of returns, and thus, the variables MARS1 (# of single returns), MARS2 (# of joint returns), and MARS4 (# of head of household returns) will all add up together to approximately equal N1. It’s not exact because in our dataset we have omitted a few other types of returns. A00100 is the adjusted group income when taken into account all of the returns in each zipcode area.

ZIP most_cell_towers
91042 398
90275 250
90012 248
91311 198
90045 160

ZIP least_cell_towers
90222 1
90506 1
90630 1
90814 1
91046 1
91607 1
91710 1
91764 1
92280 1
92305 1
92311 1
92382 1
92404 1
92518 1
92545 1
92571 1
92592 1
92612 1
92618 1
92648 1
92651 1
92676 1
ZIP least_cell_towers
92703 1
92707 1
92801 1
92806 1
92821 1
93004 1
93010 1
93030 1
93033 1
93105 1
93108 1
93205 1
93225 1
93274 1
93420 1
93513 1
93518 1
93527 1
93545 1
93549 1
94513 1
95607 1
95670 1
The zipcode with the most amount of cell towers is 91042. There are many zipcodes with only 1 cell tower, so I will not list them all since they are included in the table, but a few include 90222, 90506, and 90630. The cities in the zipcode
91042 are Tujunga, Montrose, La Crescenta, Los Angeles, Sunland, Pasadena, Mount Lukens, Valencia, Highway Highlands, Burbank, La Canada, and Sylmar.
Join data files

dealer_zip most_celltowers

90703 18031
Zipcode 90703 has the most cell towers. This new join vector has 18031 observations.

We have created a map in which we can see the locations of all of the cell towers. They are mainly concentrated in Southern California.
Analysis
t-test
I am choosing to compare the means between the MARS1 and MARS2 data from the irs dataset. Basically, I am testing to see whether there is a difference in means in the number of single returns and the number of joint returns. By looking at the difference in means, I will be able to discern which kind of tax return is more significantly filed.
Linear Regression

I have included the plot to better understand the relationship between the number of single returns and number of dependencies. In the conclusions section, I will elaborate on this plot as well as the summary output.
Custom Functions

Results & Conclusions
The p-value is 2.2 * 10^-16 which is clearly less than 0.05. This means that at the 95% significance level, there is a significant difference in means between the number of single returns and the number of joint returns. More specifically, since we ran at an alternative value of “greater,” the number of single returns is significantly greater than the number of joint returns. Thus, they are not statistically equal. Our output actually tells us the mean of the two variables. The mean of the number of single returns is 7884.340, and the mean of joint returns is 5001.771, and since they really are not values close to each other, we can be confident that our t-test worked.
The adjusted r square value is 0.5243 which means that 52.43% of the variability in the # of single returns is explained by our linear model. And honestly, that’s not too bad. This means that there is a positive, linear, moderately strong correlation between the number of dependencies and the # of single returns. This is surprising, because we should expect that if there are more dependencies in a family, for instance, then the # of single returns should decrease and instead the # of joint returns increase. That is why I believe that there is a lurking variable. The lurking variable is population. In a city, for instance, if there are many dependencies, then in that city there will be a higher number of single and joint returns. Thus, our linear model is useless since it fails to recognize this confounding variable, which has a profound effect on our conclusions.

I genuinely do not know why my function is not working, unfortunately. What a grievance this is, honestly. I have uncommented it out because the function does not let the entire html to knit otherwise. Aaah!