SACS Assessment BAS UTK
How would you visualize and summarize the overall distribution of total daily sales of one particular store? correct answersA histogram or its continuous analog (kernel density estimate) is a great way to visualize the overall distribution of a numerical quantity. Once the shape of the overall distribution is discerned, I'd also like to see a time plot to see how sales of that store have evolved over time To summarize the overall distribution, we'd use the average value of the sales (if the histogram is relatively symmetric) or the median value (if the histogram is quite skewed)
to get a "typical" value for the demand, and we'd use either the standard deviation or IQR (respectively) to get an idea of how far off values might be from this "typical" value
How would you compare (visually and statistically) sales in June between stores? correct answersA map color-coding cities (or states, depending on how you want to aggregate it)by total sales in June is a good start. Can't use a segmented bar chart or pie chart here to give the fraction of overall June sales from each store because there are too many categories (623).
A table having the "top 5" and "bottom 5" total sales in June could be useful.o Statistically, there's not much we can do. With only two Junes in the data, andthus two observations for each store, we can't compare each store statistically.Potentially you could group the stores into states or regions and compare theaverage June sales per store with an ANOVA
How would you describe (visually and statistically) how daily sales has evolved over thecourse of the study for one particular store? correct answersVisually - time plot!
Statistically - use day (1-730) as a predictor and fit a simple linear regression predicting sales from day. Check to see if the trend/slope is statisticallysignificant.
How would you examine (visually and statistically) differences in total sales for variousdays of the week? correct answersVisually - side-by-side box plots is probably the most effective. The box plot forMonday would contain sales for each store on each Monday, etc. Statistically - perform an ANOVA to see if statistically significant differences existing averages sales between the 7 days of the week (and if so, follow that up with a connecting letters report). Note: if the distribution of sales for each day is very skewed, it's probably best to compare the average value of log-Sales instead
How would you summarize how total sales in 2017 varies across the country? correct answersGeographic map where cities (or states) are color-coded by their total sales.Label the map with sales to add clarity. Statistically (if needed), could run an ANOVA to compare the average daily sales between states (probably wouldn't want to do cities...too many levels), and follow that up with a connecting letters report
How would you summarize how much of total sales comes from stores in the NW vs. SW vs. NE. vs. SE. vs. Midwest? correct answersGraphically - geographic map with regions color-coded based on fraction ofoverall sales over the 2 years that are attributed
to that region. Label withpercentages. Could also do a bar chart (with percentages labeled) or even a piechart since there aren't too many categories.
Statistically - perhaps aggregate on a week-to-week basis. For week 1 of the data, find the fraction of sales coming from the NW, from the SW, etc. Do the same for week 2, etc. Each region would then have about 104 observations, and an ANOVA could compare the average percentages between the four regions(followed up with a connecting letters report if there are differences).HOWEVER, this ignore the potential time evolution of these percentages
How would you find what combinations of items are purchased together with higher-
than-expected frequently? correct answersMarket-based analysis
How would you predict demand for a particular item (at a particular store) over the next30 days? correct answersTake BAS 475 and learn about forecasting techniques (ARIMA model or exponential smoothing is a reasonable choice) Or, fit a linear regression, random forest, etc., that looks at historical demands for that item in 30 day increments, and use as predictor variables perhaps the previously observed 30-day demand, the outdoor temperature, and anything else you could think of that could be responsible for variation in demands.
How could you detect if the type of meal (salad/burger/chicken) has an association with gender? correct answersVisually - a mosaic plot (side-by-side segmented bar charts) showing the distribution of meal type (e.g. 20% salad, 50% burger, 30% chicken) for men and for women and look for differences.
Statistically - chi-squared test of independence since we are comparing two or more categorical distributions between groups
How would you analyze whether the fraction of men who order burgers is different thanthe fraction of women who order burgers? correct answersconstruct a confidence interval for the difference in proportions of men and women who order burgers (out of 1.5 million men, 800000 order burgers; out of1.2 million women, 700000 order burgers) and see if 0 is inside (if so, no statistically significant difference)
or do a formal hypothesis test. Ho: p_men = p_women vs. HA: p_men !=p_women. Can
use Fisher's exact test since we are comparing two proportions.
How would you visualize and summarize the overall distribution of total daily sales of one particular store? correct answersA histogram or its continuous analog (kernel density estimate) is a great way to visualize the overall distribution of a numerical quantity. Once the shape of the overall distribution is discerned, I'd also like to see a time plot to see how sales of that store have evolved over time To summarize the overall distribution, we'd use the average value of the sales (if the histogram is relatively symmetric) or the median value (if the histogram is quite skewed)
to get a "typical" value for the demand, and we'd use either the standard deviation or IQR (respectively) to get an idea of how far off values might be from this "typical" value
How would you compare (visually and statistically) sales in June between stores? correct answersA map color-coding cities (or states, depending on how you want to aggregate it)by total sales in June is a good start. Can't use a segmented bar chart or pie chart here to give the fraction of overall June sales from each store because there are too many categories (623).
A table having the "top 5" and "bottom 5" total sales in June could be useful.o Statistically, there's not much we can do. With only two Junes in the data, andthus two observations for each store, we can't compare each store statistically.Potentially you could group the stores into states or regions and compare theaverage June sales per store with an ANOVA
How would you describe (visually and statistically) how daily sales has evolved over thecourse of the study for one particular store? correct answersVisually - time plot!
Statistically - use day (1-730) as a predictor and fit a simple linear regression predicting sales from day. Check to see if the trend/slope is statisticallysignificant.
How would you examine (visually and statistically) differences in total sales for variousdays of the week? correct answersVisually - side-by-side box plots is probably the most effective. The box plot forMonday would contain sales for each store on each Monday, etc. Statistically - perform an ANOVA to see if statistically significant differences existing averages sales between the 7 days of the week (and if so, follow that up with a connecting letters report). Note: if the distribution of sales for each day is very skewed, it's probably best to compare the average value of log-Sales instead
How would you summarize how total sales in 2017 varies across the country? correct answersGeographic map where cities (or states) are color-coded by their total sales.Label the map with sales to add clarity. Statistically (if needed), could run an ANOVA to compare the average daily sales between states (probably wouldn't want to do cities...too many levels), and follow that up with a connecting letters report
How would you summarize how much of total sales comes from stores in the NW vs. SW vs. NE. vs. SE. vs. Midwest? correct answersGraphically - geographic map with regions color-coded based on fraction ofoverall sales over the 2 years that are attributed
to that region. Label withpercentages. Could also do a bar chart (with percentages labeled) or even a piechart since there aren't too many categories.
Statistically - perhaps aggregate on a week-to-week basis. For week 1 of the data, find the fraction of sales coming from the NW, from the SW, etc. Do the same for week 2, etc. Each region would then have about 104 observations, and an ANOVA could compare the average percentages between the four regions(followed up with a connecting letters report if there are differences).HOWEVER, this ignore the potential time evolution of these percentages
How would you find what combinations of items are purchased together with higher-
than-expected frequently? correct answersMarket-based analysis
How would you predict demand for a particular item (at a particular store) over the next30 days? correct answersTake BAS 475 and learn about forecasting techniques (ARIMA model or exponential smoothing is a reasonable choice) Or, fit a linear regression, random forest, etc., that looks at historical demands for that item in 30 day increments, and use as predictor variables perhaps the previously observed 30-day demand, the outdoor temperature, and anything else you could think of that could be responsible for variation in demands.
How could you detect if the type of meal (salad/burger/chicken) has an association with gender? correct answersVisually - a mosaic plot (side-by-side segmented bar charts) showing the distribution of meal type (e.g. 20% salad, 50% burger, 30% chicken) for men and for women and look for differences.
Statistically - chi-squared test of independence since we are comparing two or more categorical distributions between groups
How would you analyze whether the fraction of men who order burgers is different thanthe fraction of women who order burgers? correct answersconstruct a confidence interval for the difference in proportions of men and women who order burgers (out of 1.5 million men, 800000 order burgers; out of1.2 million women, 700000 order burgers) and see if 0 is inside (if so, no statistically significant difference)
or do a formal hypothesis test. Ho: p_men = p_women vs. HA: p_men !=p_women. Can
use Fisher's exact test since we are comparing two proportions.
