Online-Data-Analyst

Notes + code (using the R statistical package) from a online data analyst job application

Link to csv file (n.b. this is dummy data):

https://www.dropbox.com/s/2worqbd25p9imx0/Online%20analyst%20test%20-%20Dummy%20data.csv?dl=0

###Work Flow / Process

1. View data variables and structure.
2. Use descriptive statistics to examine quantitative variables (using R)

###R Script

Load data

> sdata  <- read.csv("Online analyst test - Dummy data.csv", header = T)

basic overview of the variables

> str(sdata)

Visually look at the strucutre

> view(sdata)

get some descriptive statistics!

> require(psych)
> round(describe(sdata), 2)

There is a massive outlier in the variable 'sales_value' - with a skew of 502 and standard deviation of 60. Delete this outlier using MS Excel.

Create a frequency table for the catagorical variable: 'region'.

> reg.freq  <- table(sdata$region)
> reg.freq

Create a basic bar plot to examine the distribution by category for the variable 'region'.

> barplot(reg.freq, col = "lavender", main = "Barplot for Delivery Region", ylab = "Frequency")
> round(prop.table(reg.freq), 2) #just to have a quantitive look at the frequency...

Create a frequency table for the catagorical variable: 'date'.

> date.freq  <- table(sdata$Transaction_Date)
> date.freq

Create a basic bar plot to examine the distribution by category for the variable 'date'.

> barplot(date.freq, col = "lavender", main = "Barplot for Delivery Date", ylab = "Frequency")

Have a look at the descriptive statistics for the variable 'sales_value'

> describe(sdata$sales_value)

Basic plot for the variable 'sales_value'

> hist(sdata$sales_value)

As there are no further outliers value 1 SD either side of the mean equally and I can standardise the variable 'sales_value'

> sales_value.z  <- scale(sdata$sales_value)
> hist(sales_value.z)

Now that the variable is standardised I can have a look at the log distribution

> sales_value.ln  <- log(sdata$sales_value + 1)
> describe(sales_value.ln)
> hist(sales_value.ln)

Turns out that this has a lognormal distribution.

Look at the trend: formulate hypothesis', areas of opportunity and recommendations from this

Summary

• Trend wk 26: 2 ppt w/w acceleration in cart value
• Opportunities: Region 7 shows potential with a high avg. cart value
• Threats: Region 5 has seen a higher than normal drop in cart value in wk 23 - we need to understand this.

Next Steps

Create a presentation!

Points I would improve if I were to do this exercise again

• I would not use MS Excel to delete the outlier. This was an easy fix; however, Excel is not a useful tool when data becomes large and cumbersome. My edits have also not been recorded in Git. Instead I would either (1) learn how to use the ddplyr + strings package in R and split the data to remove the outlier, (2) used the data.table package, or (3), learn basic python + SQLite 3 to achieve this end. Update: I also could have R and SQLite 3 (there is the R sqldf package that I would love to learn).
• I would spend more time to analyse the time series data, including working with dates in R.
• I would use the ggplot 2 package to make my graphs. The reason being that it will be a longer term time saving solution to creating beautiful data.