An Empirical Generalization of the Effects of Category Captainship

Abstract

This paper makes use of Nielsen data to investigate the generalizability of category captainship and its impact on retailers. Category captainship is when a retailer selects a category captain to manage the product line assortment and optimization for all of the brands within a category, including the retailer’s private label and captain’s own brand. This has been shown to be optimal for the retailer and the captain and not as detrimental to the other brands as might be expected. But does this phenomenon generalize across many retailers? What characteristics attenuate the success of the practice across retailers?

Project Details

Project Description

• general effects of category captainship
• reference old paper: On the Competitive and Collaborative Implications of Category Captainship
• extend and apply across multiple chains, markets, locations and categories to get better understanding of overall effects
• Use hierarchical diff in diff model

Data

• Nielsen data, retail scanner data (RMS), from the Kilts Center archive
• weekly pricing, volume, store merchandising conditions
• 35,000 grocery, drug, mass-merchandiser, and other stores (grocery for us)
• stores are from approximately 90 retail chains
• food stores, data represents 53% of all commodity volume (ACV)
• 2011 to 2013
• North Dakota, Washington DC, Minnesota, Missouri
• Not every Nielsen retail cooperator has agreed to share their scanner data with the Kilts Center, but for retailers that do participate, typically all stores within the 48 contiguous states are included
• 3 major file types:
  • stores: individual store locations
  • products: UPC info
  • movement: price and quantity of goods sold at specific stores on a specific week
• since movement files are so large, there is one file for each product module code (category?) for each year

Working with the Data

• Data Merging/Cleaning: For each category and each year, store and movement files were merged based on store code and filtered for observations in the four areas where Supervalu Operates: North Dakota, Washington DC, Minnesota, and Missouri. Then this file was merged with the products master file based on UPC code. We then combined the merged store, movement, and product file for each year so that we had data spanning at least a year prior and a year post Category Captainship implementation. Next, we filtered out stores that switch parent or retailer codes.

Discovering Supervalu

• Graph method: There is a variable that identifies the brand for each product in our data, and we use the brand variable to tie category captains and validators to all of their products. We then filter for these products and create a new variable corresponding to each product’s manufacturer so that we can aggregate sales at the manufacturer level for each retailer. We plot aggregate sales for the category captain, validator, and store brand within each retailer. Similarly, we graph the average price level for category captain, validator, and store brand products. We expect to see increases in sales for captains and store brands post category captainship implementation, as well as significant changes in the average price level for captains and store brands. These graphs were created for the following categories for every retailer in the four Supervalu locations:

  • Ready-to-Eat Cereal
  • Spreads and Jams
  • Pickles and Olives
  • Peanut Butter
  • Novelties
  • Lunchmeat
  • Ice Cream
  • Frozen Dinners
  • Canned Soup

The graph method did not clearly indicate which retailer codes correspond to the Supervalu chains we are interested in, so we tried another method of identification.

• Table method: To discover Supervalue, we created scaled sales charts for the top 100 upcs (in terms of sales) within each retailer in each category. These charts are at the weekly level such that each row represents a upc and each cell across the rows is that week’s sales amount divided by the average sale amount for that upc over the entire time period of the data. Additionally, the cells are filled with a red color gradient that gets darker when sales are higher, and lighter (or white) when sales are lower (or zero). We expect to see significant changes (products introduced, discontinued, or large shifts in the level of sales) within a month prior and post category captainship implementation and refresh for Supervalu retailers. We have been investigating each chart for these changes. When we see something significant, we track the upc code, what company manufactures the product, and what type of change occurred. We expect to see the most changes happen for captain, validator, and private label products. We track all changes for each retailer in each category, then compare the changes across retailer codes. We believe this method has been successful in identifying Supervalu retailer codes.

Modeling

• Currently in progress

Project Organization

• /Code Scripts with prefixes (e.g., 01_import-data.R, 02_clean-data.R) and functions in /Source.
• /Data Simulated and real data, the latter not pushed.
• /Figures PNG images and plots.
• /Output Output from model runs, not pushed.
• /Presentation Presentation slides, without its knitted PDF pushed.
• /Private A catch-all folder for miscellaneous files, not pushed.
• /Writing Case studies and the paper, without its knitted PDF pushed.

For general details on collaborating using GitHub see Research Assistant Training.