22576 Chap.5 Data Enrichment and Units of Analysis
Data Enrichment and Units of Analysis
Week 4 begins by enriching cleaned data so it can answer the business question. The chapter covers feature selection to keep relevant variables, subsample analysis that compares groups, often after creating a new category from a numeric column, and composition analysis in one and two dimensions.
It then explains the unit of analysis, the level each row represents, and how aggregating from customers to stores by count, mean, median, total or rate changes what the data can show.
Enrichment decisions shape the group report and the presentation, because the analytical questions you propose depend on which variables, subsamples and levels you chose to examine.
The sample presentation questions ask exactly this, for example which new variables were created, what they added, and how very small categories were prevented from giving a misleading impression.
What this chapter covers
- 01
A systematic approach to business analytics
- 02
Feature selection and the data swamp
- 03
Subsample analysis and new categorical variables
- 04
One-dimensional and cross-tabulated composition
- 05
Choosing the unit of analysis
- 06
Aggregating by count, mean, median, total or rate
Worked example · free
Aggregate six customer visits to three store rows
- 1Counts: Store A has 3 visits, Store B has 2 and Store C has 1.
- 1Average wait: A is (4 + 6 + 8) ÷ 3 = 6.0 minutes, B is (5 + 9) ÷ 2 = 7.0 and C is 10.0.
- 1Average satisfaction: A is 23 ÷ 3 = 7.67, B is 16 ÷ 2 = 8.0 and C is 5.0.
Key terms
- Feature selection
- Choosing the variables most relevant to a business problem to cut noise and improve model performance.
- Composition analysis
- Describing how observations are distributed across groups, as counts or shares that sum to 100%.
- Subsample analysis
- Examining defined subsets of the data to compare groups and expose hidden patterns.
- Aggregation
- Combining rows into a higher-level unit using a count, mean, median, total or rate.
Data Enrichment and Units of Analysis FAQ
What is the unit of analysis?
It is the level at which a question is answered and what each row of the data represents, such as a customer, a transaction, a store or the whole firm. The same question about satisfaction can be asked at each level, and each gives different patterns and conclusions.
What is lost when data are aggregated?
Detail below the new unit disappears. Combining visits into one row per store removes visit-level variables, such as the day of the week, and averages can hide a very small group behind a single number. Choose the aggregation for each column to suit the question.
How do I create a new categorical variable for subsamples?
Cut a numeric column into bands that cover every value without overlapping, such as budget, mid-range and premium price bands. The new variable lets you compare groups, for example average ratings by price band, which the original continuous column cannot show directly.
What is the difference between one-dimensional and cross-tabulated composition?
A one-dimensional composition shows how observations split across the categories of one variable, such as age bands. A cross-tabulated composition shows how that split changes across a second variable, such as product shares within each region, with each row adding to 100%.
Why does feature selection matter before modelling?
Keeping only relevant variables defines the scope of the analysis, cuts noise, speeds up modelling and avoids including several variables that carry the same information. Correlation analysis from Week 3 is the main tool for spotting those overlapping variables and keeping one from each group.
How does composition analysis help a report?
It shows readers what the sample is made of before they read any result, such as the mix of age groups or regions. That context reveals whether the sample is balanced and representative, and it can expose an over-represented group that should be resampled before groups are compared.
Assessment move
For each dataset you meet, write one sentence naming its current unit of analysis and one naming the unit a manager's question needs. Practise the GroupBy and Pivot nodes until you can build a store-level table in a couple of minutes. Practise writing the same business question at three levels, customer, store and firm, and list the columns each version would need.
Working through Data Enrichment and Units of Analysis in 22576? Sia is AskSia’s AI Business Analytics tutor — ask any 22576 Data Enrichment and Units of Analysis question and get a clear, step-by-step explanation grounded in how 22576 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
