22576 Chap.9 Linear Regression: Building and Interpreting Models
Linear Regression: Building and Interpreting Models
Week 6 begins supervised learning with linear regression for continuous outcomes. This chapter covers the basic and multi-variable equations, reading coefficients, intercepts and p-values, and predicting and computing residuals.
It explains dummy variables with a reference group, why a coefficient in a multi-variable model holds other variables constant, why a variable can lose significance once controls are added, and how R-squared and mean absolute error describe fit.
The worked examples read real output tables line by line, because quiz items and the individual report both reward precise interpretation more than the clicks that produce the output.
The chapter also flags the habits that cost marks: reading a dummy without naming its reference group, dropping the 'holding other variables constant' clause, and treating a large but insignificant coefficient as a finding.
What this chapter covers
- 01
Predictive modelling with X inputs and a Y outcome
- 02
When linear regression suits the outcome
- 03
Basic and multi-variable regression equations
- 04
Interpreting coefficients, intercepts and p-values
- 05
Dummy variables and the reference group
- 06
R-squared, adjusted R-squared, residuals and MAE
Worked example · free
Read channel dummies against an Email reference group
- 1Email is the reference group, so its predicted average is the intercept: $52.
- 1SMS: 52 + 9 = $61, which is $9 more than Email, significant at the 5% level but not at 1%.
- 1App push: 52 + 21 = $73, which is $21 more than Email, significant at the 1% level.
- 1Each dummy coefficient is a difference from Email, not an average on its own.
Key terms
- Regression coefficient
- The expected change in the outcome for a one-unit increase in an explanatory variable.
- Dummy variable
- A 0/1 column representing one category of a categorical variable in a regression.
- R-squared
- The share of variation in the outcome that a regression model explains, from 0 to 1.
- Mean absolute error
- The average size of prediction errors, in the units of the outcome.
- Ceteris paribus
- Holding all other variables constant, the condition attached to a multi-variable coefficient.
- Residual
- The difference between an observation's actual value and the value the model predicts.
Linear Regression: Building and Interpreting Models FAQ
Why must one category be left out of a dummy regression?
The excluded category is the reference group that the others are measured against; the intercept describes it and each dummy coefficient is a difference from it. If you let KNIME choose, it picks the reference itself, so the tutorial builds dummies explicitly and leaves out the chosen baseline.
Why can a variable lose significance when controls are added?
Because it may have been standing in for something the controls now capture. In the Airbnb tutorial, guest numbers mattered on their own, but once bedrooms and room type entered the model they absorbed the effect of property size, and guest numbers were no longer significant.
What does an R-squared of 0.15 mean?
The model explains about 15% of the variation in the outcome, so other factors drive most of it. A variable can still have a significant coefficient with a low R-squared; significance concerns that one effect, while R-squared concerns how much of the outcome the whole model accounts for.
When is linear regression the wrong model?
When the outcome is categorical, such as whether a customer complains, a classification model like logistic regression is needed. When the outcome is a discrete count, the workshop points to tree-based models. Linear regression suits continuous outcomes such as revenue, turnover or return on investment.
What is a residual?
It is the difference between the actual value of the outcome and the value the model predicts for that observation, actual minus predicted. Every observation has one; smaller residuals overall mean a more accurate model, and their pattern is used to check the regression assumptions.
What does the intercept mean in a regression?
It is the predicted value of the outcome when every explanatory variable equals zero. In a dummy regression it is the average for the reference group. When zero lies outside the observed data, such as zero advertising for a large retailer, the intercept is a mathematical anchor and should be interpreted cautiously.
Assessment move
For every regression output you meet, write three sentences: the equation, the key coefficient with direction, size, units and significance, and the fit. Practise dummy interpretations aloud, always naming the reference group, until the wording is automatic. Rebuild the Week 7 Airbnb workflow, then explain to a classmate why guest numbers lost significance once bedrooms entered the model.
Working through Linear Regression: Building and Interpreting Models in 22576? Sia is AskSia’s AI Business Analytics tutor — ask any 22576 Linear Regression: Building and Interpreting Models 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.
