📘 Lecture 6 — Demand Sensitivity Analysis: The Elasticity Concept
📖 Overview: This lecture introduces the concept of elasticity as a measure of responsiveness in demand analysis. It covers price elasticity of demand, its calculation methods (point and arc elasticity), the relationship between elasticity, marginal revenue, and total revenue, as well as cross-price and income elasticities—all essential for optimal pricing and managerial decision-making.
🗂️ Topics Covered
The lecture covers the elasticity concept and its measurement through point and arc elasticity, the optimal price formula using MR=MC, the relationship between marginal revenue, total revenue, and price elasticity, varying elasticity along a linear demand curve, constant price elasticity of demand, factors affecting price elasticity of demand, cross-price elasticity of demand, income elasticity of demand (normal vs. inferior goods), and the use of elasticities in managerial decision-making with a regression example.
📝 Lecture Summary
THE ELASTICITY CONCEPT
For useful managerial decision making, the firm must know the sensitivity or responsiveness of demand to changes in factors that make up the underlying demand function. One measure of responsiveness employed not only in demand analysis but throughout managerial decision making is elasticity, defined as the percentage change in a dependent variable, Y, resulting from a 1 percent change in the value of an independent variable, X. The equation for calculating elasticity is: E = %Δ Quantity / %Δ Price.
POINT ELASTICITY AND ARC ELASTICITY
Elasticity can be measured in two different ways: point elasticity and arc elasticity. Point elasticity shows sensitivity of Y to small changes in X. The most widely used elasticity measure is the price elasticity of demand, which measures the responsiveness of the quantity demanded to changes in the price of the product, holding constant the values of all other variables in the demand function.
🔑 Definition — Point Elasticity: εₓ = ∂Y/Y ÷ ∂X/X (sensitivity to small changes).
Arc elasticity shows sensitivity of Y to big changes in X. Eₓ = ΔY/ΔX * (X₂+X₁)/(Y₂+Y₁).
Elasticity Varies along Demand Curve:
- As price rises, so too does │εₚ│.
- As price falls, so too does │εₚ│.
- In all cases, εₚ < 0.
OPTIMAL PRICE FORMULA
Price elasticity estimates represent vital information because these data, along with relevant unit cost information, are essential inputs for setting a pricing policy that is consistent with value maximization. There is a relatively simple mathematical relation between marginal revenue, price, and the point price elasticity of demand.
Starting from TR = P*Q, applying the product rule: MR = dTR/dQ = P + Q * dP/dQ = P + P[Q/P * dp/dQ] = P + P(1/E). Therefore: MR = P[1 + (1/E)]. Profit maximization requires MR = MC, so P[1 + (1/E)] = MC.
🔑 Formula — Optimal Price: P* = MC / [1 + (1/εₚ)] → The optimal price equals marginal cost divided by one plus the reciprocal of price elasticity.
📌 Example: If MC = $10 and εₚ = -3, then P* = 10 / [1 + (1/(-3))] = 10 / [1 - 0.333] = 10 / 0.667 = $15.
MARGINAL REVENUE, TOTAL REVENUE, AND PRICE ELASTICITY
There are simple, direct relations between price elasticity, marginal revenue, and total revenue. The relationship between price and revenue depends on elasticity. A price fall will reduce receipts, but because the demand curve is downward sloping, the drop in price will also increase quantity demanded. Depending on the degree of price elasticity, a reduction in price can increase, decrease, or leave total revenue unchanged.
- Price cut increases revenue if │εₚ│ > 1 (elastic demand).
- Revenue constant if │εₚ│ = 1 (unitary elasticity).
- Price cut decreases revenue if │εₚ│ < 1 (inelastic demand).
ELASTICITY VARIES ALONG A LINEAR DEMAND CURVE
All linear demand curves, except perfectly elastic or perfectly inelastic ones, are subject to varying elasticities at different points on the curve. A linear demand curve is price elastic at some output levels but inelastic at others.
- Relative elasticity of demand: Eₚ > 1
- Relative inelasticity of demand: 0 < Eₚ < 1
- Unitary elasticity of demand: Eₚ = 1
- Perfect elasticity: Eₚ = ∞
- Perfect inelasticity: Eₚ = 0
As price decreases: revenue rises when demand is elastic, revenue falls when it is inelastic, and revenue reaches its peak if elasticity = 1.
If P = AR = a - bQ, then TR = aQ - bQ² and MR = a - 2bQ. Marginal revenue is positive in the range where demand is price elastic, zero where Eₚ = -1, and negative in the inelastic range.
CONSTANT PRICE ELASTICITY OF DEMAND
Some demand curves have constant elasticity; the demand curve assumes the shape of a rectangular hyperbola (so that TR is constant regardless of price). Such a curve has a nonlinear equation: Q = aP⁻ᵇ, where –b is the elasticity coefficient and equals –b throughout the demand curve.
FACTORS AFFECTING THE PRICE ELASTICITY OF DEMAND
There are three major influences on price elasticities: (1) the extent to which a good is considered to be a necessity; (2) the availability of substitute goods to satisfy a given need; and (3) the proportion of income spent on the product.
The size of the price elasticity of demand is larger the closer and the greater is the number of available substitutes for the commodity. The more narrowly a commodity is defined, the greater is its price elasticity of demand because the greater will be the number of substitutes. Similarly, the demand for "big ticket" items (automobiles, homes) accounts for a large share of consumer income and will be relatively sensitive to price, while demand for less expensive products (soft drinks, candy) can be relatively insensitive to price.
CROSS-PRICE ELASTICITY OF DEMAND
We can measure the responsiveness in the demand for commodity X to a change in the price of commodity Y with the cross-price elasticity of demand (Eₓᵧ). This is given by the percentage change in the demand for commodity X divided by the percentage change in the price of commodity Y, holding constant all other variables.
🔑 Definition — Cross-Price Elasticity: εₚₓ = ∂Qᵧ/Qᵧ ÷ ∂Pₓ/Pₓ = ∂Qᵧ/∂Pₓ * Pₓ/Qᵧ.
If the value of Eₓᵧ is positive, commodities X and Y are substitutes (examples: coffee and tea, butter and margarine, Coca-Cola and Pepsi). If Eₓᵧ is negative, commodities X and Y are complements (examples: coffee and sugar, cars and gasoline). If Eₓᵧ is close to zero, X and Y are independent commodities.
Firms use cross-price elasticity to measure the effect of changing the price of one product on the demand of other related products that the firm also sells.
INCOME ELASTICITY OF DEMAND
The income elasticity of demand measures the responsiveness of demand to changes in income, holding constant the effect of all other variables that influence demand. Income and the quantity purchased typically move in the same direction.
🔑 Definition — Income Elasticity: εᵢ = ∂Q/Q ÷ ∂I/I.
Normal goods have εᵢ > 0 (income and demand move in the same direction). Inferior goods have εᵢ < 0 (demand declines as income increases, e.g., beans and potatoes).
One important use of income elasticity is in forecasting the change in demand under different economic conditions. Demand for a commodity with low-income elasticity will not fluctuate much during booms or recessions, while demand for a luxury item will increase during booms and fall sharply during recessions.
USING ELASTICITIES IN MANAGERIAL DECISION MAKING
Elasticities are used in production and cost analysis. Factors within the control of the firm (price, advertising) are called endogenous variables. Factors outside the control of the firm (consumer incomes, competitor prices, weather) are called exogenous variables. The effects of changes in both types of influences must be understood for effective decision making.
📌 Example Problem: ABC Company markets coffee brand X with the estimated demand function: Qₓ = 1.5 - 3.0Pₓ + 0.8I + 2.0Pᵧ - 0.6Pₛ + 1.2A
Given: Pₓ = $2, I = $2.5 trillion, Pᵧ = $1.80, Pₛ = $0.50, A = $1 (hundred thousands)
Step 1: Calculate Qₓ = 1.5 - 3(2) + 0.8(2.5) + 2(1.80) - 0.6(0.50) + 1.2(1) = 2 million pounds
Step 2: Calculate elasticities:
- Eₚ = -3[2/2] = -3 (price elasticity)
- Eᵢ = 0.8[2.5/2] = 1 (income elasticity - normal good)
- Eₓᵧ = 2[1.8/2] = 1.8 (cross-price with competitor - substitute)
- Eₓₛ = -0.6[0.5/2] = -0.15 (cross-price with sugar - complement)
- Eₐ = 1.2[1/2] = 0.6 (advertising elasticity)
💡 Why this matters: These elasticity estimates allow the firm to predict exactly how changes in price, income, competitor actions, and advertising will affect sales, enabling optimal pricing and marketing decisions.
⭐ Key Takeaways
The most critical concept from this lecture is that elasticity measures the percentage responsiveness of demand to changes in various factors, and it is essential for pricing decisions through the optimal price formula P* = MC/[1+(1/εₚ)]. Students must understand the three elasticity ranges (elastic, unitary, inelastic) and their effect on total revenue when price changes. The distinction between point elasticity (for small changes) and arc elasticity (for large changes) is crucial, as is the recognition that elasticity varies along a linear demand curve. Cross-price elasticity (positive for substitutes, negative for complements) and income elasticity (positive for normal goods, negative for inferior goods) are equally important for comprehensive demand analysis. Finally, the ability to compute and interpret elasticities from estimated demand functions is a practical skill required for managerial decision-making.
🧠 Quick Revision Questions
-
A firm's demand has price elasticity of -0.8. If the firm lowers its price, what will happen to total revenue? Explain why.
-
Given MC = $12 and εₚ = -4, calculate the optimal price using the optimal price formula.
-
If the cross-price elasticity between Product A and Product B is +2.5, what type of relationship exists between these products? What happens to the demand for Product A if the price of Product B rises by 10%?
-
A luxury car company estimates income elasticity of demand for its vehicles at +2.8. During a recession when incomes fall by 5%, by what percentage would demand change?
-
On a linear demand curve, where is marginal revenue equal to zero, and what is the value of price elasticity at that point?
📘 Lecture 7 — DEMAND ESTIMATION
📖 Overview: This lecture introduces the fundamental methods of estimating demand for a product or service, moving from conceptual problems like the identification problem to practical approaches like regression analysis and marketing research. Understanding these estimation techniques is crucial for managers to make informed pricing, production, and marketing decisions based on empirical data.
🗂️ Topics Covered
The lecture begins by addressing the simple demand curve estimation and the critical "identification problem" that arises from the interplay of demand and supply shifts. It then explores three key marketing research approaches: consumer interviews/surveys, consumer clinics, and market experiments. Finally, it provides a comprehensive introduction to regression analysis, distinguishing between deterministic and statistical relations, and explaining data types like time series and cross-section data, along with the use of scatter diagrams.
📝 Lecture Summary
SIMPLE DEMAND CURVE ESTIMATION
For simple Linear Demand Curves, the best estimation method balances marginal costs and marginal benefits. This approach is useful because straight-line relations can often provide helpful approximations in demand estimation.
🔑 Definition — Simple Linear Demand Curve Estimation: A method that uses straight-line relations to approximate demand, balancing the marginal costs and marginal benefits of estimation.
THE IDENTIFICATION PROBLEM
The demand curve for a commodity is generally estimated from market data on the quantity purchased at various prices over time (time series data) or for various consuming units at one point in time (cross-section data). However, simply joining price-quantity observations on a graph does not generate the true demand curve because each observation is given by the intersection of a different, unobserved demand and supply curve. The main problems are:
- Changing Nature of Demand Relations: Demand relations are dynamic.
- Interplay of Demand and Supply: Economic conditions affect both.
- Shifts in Demand and Supply: Curve shifts must be estimated.
- Simultaneous Relations: Quantity and price are jointly determined, causing the “Simultaneity Problem” where resulting equilibriums do not trace out supply or demand.
To derive the true demand curve from observed data points, we must allow the supply curve to shift freely while adjusting for shifts in the demand curve caused by changes in consumers' incomes, prices of related commodities, tastes, and other factors. This adjustment isolates the effect on quantity demanded resulting only from a change in the commodity's price.
By including the most important determinants of demand as independent variables, regression analysis allows the researcher to unravel the independent effects of various determinants, isolating the price effect to identify the demand curve. It is the uncorrected shifts in supply, after adjusting for demand shifts, that allow us to derive a particular demand curve. For example, in the lecture's figure, point E’ on demand curve D₂ is derived by correcting shifts in demand while allowing the supply curve to shift from S₂ to S₁.
💡 Why this matters: The identification problem is a core challenge in econometrics; if not addressed, estimated demand curves will be biased and useless for predicting consumer response to price changes.
MARKETING RESEARCH APPROACH TO DEMAND ESTIMATION
Although regression analysis is the most useful method, marketing research approaches are also used. The most important of these are:
- Consumer Interviews (or surveys)
- Consumer Clinic
- Market Experiments
CONSUMER INTERVIEWS OR SURVEYS
Surveys involve questioning a sample of consumers about their likely responses to changes in price, income, the price of related commodities, advertising, credit incentives, and other demand determinants. These surveys can be conducted informally or with sophisticated questionnaires administered to a representative sample by trained interviewers.
There are two types of questions used in questionnaires:
- Specific and closed questions: Used to obtain specific information, seeking "Yes/No" answers or choices among alternatives. For example, KFC and McDonald's use a Likert scale of 5 to rate categories like food quality, cleanliness, service, atmosphere, and staff behavior, with choices: Excellent, Good, Fairly good, Satisfactory, Poor.
- Open-ended questions: Allow respondents to answer in their own way, designed to encourage extensive and developmental answers to reveal attitudes and facts.
In theory, consumer questionnaires can provide useful information, but they are often biased because consumers are unable or unwilling to provide accurate answers, especially regarding personal information like age, income, and taxes.
CONSUMER CLINICS
Consumer clinics are laboratory experiments where participants are given money and asked to spend it in a simulated store. This allows researchers to observe reactions to changes in price, packaging, displays, and competing product prices. Participants are selected to represent the target market's socioeconomic characteristics, and they can keep the purchased goods, providing an incentive for realistic behavior.
Consumer clinics are more realistic than surveys but have serious shortcomings. The results are questionable because participants know they are in an artificial situation and being observed, so they may not act normally as they would in a real market.
MARKET EXPERIMENTS
Unlike consumer clinics, market experiments are conducted in the actual marketplace. One method involves selecting several markets with similar socioeconomic characteristics and changing the price, packaging, or promotion in different markets, then recording consumer purchases. Census data or surveys can also determine the effect of age, gender, education, income, and family size on demand.
The advantages of market experiments are that they can be conducted on a large scale for validity, and consumers are unaware they are part of an experiment. However, they have serious disadvantages:
- They may be conducted on too limited a scale and over too short a period.
- Extraneous occurrences (e.g., a strike or bad weather) may bias results.
- Competitors could sabotage or monitor the experiment.
- A firm might permanently lose customers in markets where it experiments with high prices.
Despite these shortcomings, market experiments are useful for determining pricing strategy, testing packaging and promotional campaigns, especially for new products with no existing data.
REGRESSION ANALYSIS
To understand the use of regression analysis, one must distinguish between two broad classes of economic relations:
- A deterministic relation is known with certainty. For example, total profit equals total revenue minus total cost (π = TR – TC). Once TR and TC are known, total profits can be exactly determined.
- A statistical relation exists between two variables if the average of one is related to another, but it is impossible to predict the exact value. For example, if TC = $10Q on average, a one-unit increase in Q tends to result in an average $10 increase in TC, but the actual increase may be more or less.
When a statistical relation exists, the true relation is unknown and must be estimated, typically by gathering and analyzing historical data.
Two types of data are commonly used:
- Time series of data: A daily, weekly, monthly, or annual sequence of data on an economic variable (e.g., price, income, cost). This helps judge trends over time.
- Cross section of data: A group of observations on an economic variable at any point in time. This helps analyze the relative importance of factors like market share vs. advertising as determinants of profitability.
The simplest way to analyze a sample of historical data is to plot a scatter diagram, which plots the dependent variable on the Y-axis and the independent variable on the X-axis. The lecture shows scatter diagrams plotting demand against four factors:
- Figure (a) depicts an inverse relation between quantity sold and price.
- Figure (b) shows a direct relation between advertising and demand.
- Figure (c) shows no relation between demand and the price of an unrelated product (Product X).
- Figure (d) illustrates a nonlinear relation between demand and income.
Scatter diagrams are analyzed to gain an instinctive "feel" for the data, using an inductive and intuitive method. Although valuable as a starting point, their lack of structure can limit their value.
💡 Why this matters: Regression analysis provides the statistical framework to move beyond visual inspection of scatter plots and quantify the precise relationships between demand and its determinants, enabling reliable forecasting.
⭐ Key Takeaways
The lecture emphasizes that estimating demand is not straightforward due to the identification problem, where price and quantity data are jointly determined by shifting demand and supply. To overcome this, one must use regression analysis that controls for other demand shifters while allowing supply to vary. Three non-statistical methods—consumer surveys, consumer clinics, and market experiments—offer useful but imperfect alternatives, each with biases from artificial settings or limited scope. The choice between deterministic and statistical relations is fundamental: demand estimation always involves statistical relations due to random variation. Finally, the first step in any data analysis is creating scatter diagrams to visually explore relationships, forming the basis for more rigorous regression analysis.
🧠 Quick Revision Questions
- What is the "identification problem" in demand estimation, and why does simply plotting price-quantity data points fail to reveal the true demand curve?
- List the three marketing research approaches to demand estimation and state one major advantage and one major disadvantage of each.
- Distinguish between a "deterministic relation" and a "statistical relation," and provide an example of each from the lecture.
- What is the difference between time series data and cross-section data, and when might a firm use each type?
- Using a scatter diagram, describe the expected relationship between the quantity demanded and (a) the price of the good, (b) advertising expenditure, and (c) the price of an unrelated good.
📘 Lecture 8 — Demand Estimation (Continued 1)
📖 Overview: This lecture introduces regression analysis as a statistical technique for estimating demand relationships. It covers the steps involved in regression analysis, functional form specifications, the ordinary least squares (OLS) method, and how to interpret and test the statistical significance of regression results. Understanding regression analysis is crucial for managers to quantify relationships between variables like advertising and sales, enabling data-driven decision-making.
🗂️ Topics Covered
The lecture begins with the historical origin of the term "regression" and its modern interpretation. It then outlines the steps in regression analysis including variable specification, data collection, functional form selection, parameter estimation, and result interpretation. The two main functional forms—linear and multiplicative (power)—are presented. A detailed introduction to regression analysis uses an advertising-sales example with a scatter diagram. The ordinary least squares (OLS) model is explained, including the concept of minimizing squared errors. Finally, the lecture covers tests of significance using the t-statistic and the calculation of confidence intervals for the estimated parameter.
📝 Lecture Summary
REGRESSION ANALYSIS
The term "regression" was first used by geneticist Francis Galton (1886), who observed that tall fathers have shorter sons and short fathers have taller sons. He described this as regression where height tended to "regress" towards the mean height for the population.
THE MODERN INTERPRETATION OF REGRESSION
Regression Analysis is a powerful statistical technique that describes the way in which one important economic variable is related to one or more other economic variables.
STEPS IN REGRESSION ANALYSIS
- Specify variables: Quantity Demanded, Advertising, Income, Price, Other prices, Quality, Previous period demand.
- Obtain data: Cross sectional vs Time series.
- Specify functional form of equation:
- Linear:
Yt = a + b X1t + g X2t + ut - Multiplicative:
Yt = a X1t^b X2t^g e^twhich is estimated asln Yt = ln a + b ln X1t + g ln X2t + ut
- Linear:
- Estimate parameters
- Interpret results: economic and statistical.
FUNCTIONAL FORM SPECIFICATIONS
Linear Function (Linear Model):
- Estimation Format:
QX = a0 + a1 PX + a2 I + a3 N + a4 PY + ... + e
Power Function:
- Form:
Q = a (P^b1)(P^b2) - Estimation Format:
ln QX = ln a + b1 ln PX + b2 ln PY
INTRODUCTION TO REGRESSION ANALYSIS
To introduce regression analysis, suppose a manager wants to determine the relationship between the firm's advertising expenditures and its sales revenue. The manager hypothesizes that higher advertising leads to higher sales and wants to estimate the strength of this relationship. Data on advertising and sales for the past 10 years is collected. The level of advertising expenditures (X) is the independent or explanatory variable, while sales revenues (Y) is the dependent variable.
Advertising Expenditures and Sales Revenues of the Firm (in millions of dollars)
| Year | X | Y |
|---|---|---|
| 1 | 10 | 44 |
| 2 | 9 | 40 |
| 3 | 11 | 42 |
| 4 | 12 | 46 |
| 5 | 11 | 48 |
| 6 | 12 | 52 |
| 7 | 13 | 54 |
| 8 | 13 | 58 |
| 9 | 14 | 56 |
| 10 | 15 | 60 |
A scatter diagram is a plot of each pair of (X, Y) values on a graph. From the scatter diagram for this data, a positive and approximately linear relationship is observed between advertising and sales.
A line can be visually fitted to these points. The difficulty with visual fitting is subjectivity; different researchers would get different results. Regression analysis is a statistical technique for obtaining the line that best fits the data points according to an objective statistical criterion, so that all researchers get the same result. The regression line minimizes the sum of the squared vertical deviations of each point from the regression line. This method is called the "ordinary least-squares method" (OLS).
In the figure, Y1 is the actual or observed sales revenue ($44M) for the first year. Ŷ1 (Y hat sub 1) is the estimated sales revenue from the regression line for the same advertising expenditure. The symbol e1 is the vertical deviation or error of the actual value from the estimated value for the first year.
ORDINARY LEAST SQUARES (OLS) MODEL
The OLS model is specified as:
Yt = a + b Xt
Ŷt = â + b̂ Xt + et
et = Yt - Ŷt
The regression line minimizes the sum of the squared vertical deviations (errors), i.e., Σe^2.
Errors arise because:
- Various explanatory variables are absent.
- Possible errors of measurement in Y.
- Random human behavior that leads to different results under identical conditions.
Simple Regression Analysis includes:
- Calculate the value of a and b.
- Tests of significance of parameter estimates.
- Confidence interval for the true parameter.
- Overall explanatory power of the regression.
The objective of OLS is to determine the slope and intercept that minimize the sum of the squared errors. The objective is:
Min Σet² = Σ(Yt - Ŷt)² = Σ(Yt - â - b̂Xt)²
The value of b̂ (estimated slope) is given by:
b̂ = Σ( Xt - X' )( Yt - Y' ) / Σ( Xt - X' )²
Where Y' and X' are the mean or average values of Y and X, respectively.
The value of â (estimated intercept) is then obtained from:
â = Y' - b̂ X'
Calculation Example:
Given n = 10, ∑Xt = 120, ∑Yt = 500, ∑(Xt - X')² = 30, ∑(Xt - X')(Yt - Y') = 106, and ∑et² = 65.4830.
X' = 120/10 = 12Y' = 500/10 = 50b̂ = 106 / 30 = 3.533â = 50 - (3.533)(12) = 7.60Thus, the estimated regression line is: Ŷ = 7.60 + 3.533 X
The standard error of the slope (s_b̂) is calculated as:
s_b̂ = √[ Σ(et)² / ((n – k) * Σ( Xt – X' )²) ]
Where k is the number of parameters estimated (2 for slope and intercept).
s_b̂ = √[ 65.4830 / ((10 – 2) * 30) ] = √[ 65.4830 / 240 ] = √0.2728 = 0.52
TESTS OF SIGNIFICANCE: CALCULATION OF THE T STATISTIC
The t-statistic (or t-ratio) is used to test the hypothesis that the true value of b is zero (i.e., no relationship between X and Y). It is calculated as:
t = b̂ / s_b̂
For the example:
t = 3.53 / 0.52 = 6.79
Degrees of Freedom = (n - k) = (10 - 2) = 8
To conduct a significance test, the calculated t-ratio is compared to the critical value from the t-distribution table. For a two-tailed test at the 5% level of significance with 8 degrees of freedom, the critical value is 2.306.
Since the calculated value of t = 6.79 exceeds the critical value of t = 2.306, the null hypothesis (that there is no relationship between X and Y) is rejected. The alternative hypothesis (that there is a significant relationship) is accepted. This means we are 95% confident that a relationship exists.
CONFIDENCE INTERVAL FOR THE TRUE B
Confidence intervals can also be calculated for the true value of b. A 95% confidence interval is:
b̂ ± t_critical * (s_b̂)
Where t_critical is the value for the 5% level of significance.
For the example:
3.53 ± 2.306 (0.52) → 3.53 ± 1.20
🔑 This means: We are 95% confident that the true value of b lies between 2.33 and 4.73.
A 99% confidence interval uses the t-critical value for 1% level of significance (0.01), which for 8 df is 3.355.
3.53 ± 3.355(0.52)
🔑 This means: We are 99% confident that the true value of b lies between 1.79 and 5.27.
⭐ Key Takeaways
Regression analysis is an objective statistical technique that describes the relationship between a dependent variable (like sales) and one or more independent variables (like advertising). The Ordinary Least Squares (OLS) method estimates this relationship by minimizing the sum of squared errors, producing an estimated intercept (â) and slope (b̂). The statistical significance of the estimated slope is tested using a t-statistic, which is compared to a critical value; if the calculated t exceeds the critical value, we reject the null hypothesis of no relationship. Finally, confidence intervals provide a range within which the true population parameter is likely to fall with a given level of confidence (e.g., 95% or 99%).
🧠 Quick Revision Questions
- What is the key difference between visually fitting a line to data and using Ordinary Least Squares (OLS)?
- For the advertising-sales example, calculate the estimated sales revenue (Ŷ) when advertising expenditure (X) is 12.
- A t-statistic of 1.5 is calculated with 8 degrees of freedom. Can you reject the null hypothesis at the 5% level of significance? (Critical value = 2.306).
- If the estimated slope (
b̂) is 3.53 and its standard error is 0.52, what is the 95% confidence interval for the true slope? - What are the two functional forms for a demand equation mentioned in the lecture, and what is the estimation format for the multiplicative (power) function?
📘 Lecture 9 — Demand Estimation (Continued 2)
📖 Overview: This lecture continues the study of demand estimation by diving deep into regression analysis. It covers the critical assumptions underlying regression, how to specify a regression model, and how to evaluate its explanatory power using key statistical measures like R², adjusted R², and the F-statistic. Understanding these concepts is essential for managers to reliably estimate demand functions and make data-driven decisions.
🗂️ Topics Covered
The lecture begins by outlining the core assumptions of regression analysis, including normality, homoscedasticity, and no autocorrelation. It then explains the process of specifying a regression model, from variable selection to choosing a functional form (linear vs. multiplicative). The main body focuses on the coefficient of determination (R²) as a measure of goodness of fit, including a detailed calculation example. It also covers the coefficient of correlation, the multiple regression model (with an example including advertising and quality control), the adjusted R², the F-statistic for overall significance, and common problems in regression like multicollinearity, heteroskedasticity, and autocorrelation, including the Durbin-Watson statistic for testing autocorrelation.
📝 Lecture Summary
ASSUMPTIONS OF REGRESSION ANALYSIS
Regression analysis is built on several key assumptions about the error terms (residuals). First, they should be normally distributed. Second, they must have a zero expected value or mean. Third, they must have constant variance in each time period and for all values of X, a condition known as homoscedasticity (equal variance). Fourth, the value of the error term in one time period must be unrelated to its value in another period, meaning there is no autocorrelation between any two error terms. Finally, there must be no perfect multicollinearity, meaning no perfect linear relationship exists among the explanatory variables.
🔑 Definition — Homoscedasticity: The condition where the variance of the error term is constant for all values of the independent variable(s). 🔑 Definition — Multicollinearity: A situation in regression analysis where two or more independent variables are highly linearly correlated, which can make it difficult to estimate the individual effect of each variable on the dependent variable.
SPECIFYING THE REGRESSION MODEL
The first step in regression analysis is to specify the variables. The dependent variable is product demand (e.g., in physical units). The independent variables include the price of the product, prices of related goods (complements and substitutes), advertising expenditures, consumer incomes, and population. For expensive durable goods, variables like interest rates and credit terms are included.
The second step is to obtain reliable data. The third step is to determine the functional form of the regression equation. The most common form is the linear model, where the effect of each independent variable is assumed to be additive and constant. 📐 Formula: Q = a + bP + cA + dI → Q = unit demand, P = price, A = advertising, I = per capita disposable income. Here, 'b' is the estimated change in Q for a one-unit change in P, holding A and I constant.
Another common form is the multiplicative model (or log-linear model), used when the marginal effect of one variable depends on the values of other variables. 📐 Formula: Q = aPᵇAᶜIᵈ → This model implies constant elasticity, meaning the percentage change in Q due to a one percent change in an independent variable (like P) is constant. 💡 Why this matters: The choice between a linear and multiplicative model depends on the underlying economic relationship. A multiplicative model is often more realistic for demand functions.
🔑 Definition — Constant Elasticity: A property of the multiplicative (log-linear) regression model where the elasticity of Y with respect to an X variable is a constant value, regardless of the level of X.
TEST OF GOODNESS OF FIT AND CORRELATION
The coefficient of determination, denoted by R², measures the overall explanatory power of the entire regression. It is defined as the proportion of the total variation in the dependent variable (Y) that is explained by the variation in the independent variable(s). This is calculated by decomposing the total variation in Y into explained and unexplained parts.
🔑 Definition — Coefficient of Determination (R²): The proportion of the total variation in the dependent variable (Y) that is explained by the variation in the independent variable(s) in the regression. 📐 Formula: Total Variation = Explained Variation + Unexplained Variation ∑(Yₜ - Ȳ)² = ∑(Ŷₜ - Ȳ)² + ∑(Yₜ - Ŷₜ)² → Ȳ is the mean of Y, Ŷₜ is the predicted value from the regression. 📐 Formula: R² = Explained Variation / Total Variation = 1 - (Unexplained Variation / Total Variation)
📌 Example: Using the advertising-sales data from the lecture table (n=10 years): Total variation in sales (Y) = 440.0 Explained variation in sales (Y) = 373.843 Unexplained variation in sales (Y) = 65.483 Therefore, R² = 373.84 / 440.00 = 0.85 This means 85% of the total variation in the firm's sales is accounted for by the variation in its advertising expenditures.
The coefficient of correlation, denoted by 'r', is the square root of R². It measures the degree of association or covariation between X and Y. It ranges from -1.00 to 1.00, with values close to ±1 indicating a strong relationship. 📐 Formula: r = √R² 📌 Example: For the advertising-sales example, r = √0.85 = 0.92. This means sales and advertising vary together 92% of the time.
THE MULTIPLE REGRESSION MODEL
When the dependent variable is hypothesized to depend on more than one independent variable, we use multiple regression analysis. The model can be written as: 📐 Formula: Y = a + b₁X₁ + b₂X₂ + ... + b'ₖXₖ Where Y is the dependent variable, and X₁, X₂,...Xₖ are the independent variables. The only additional assumptions beyond those for simple regression are that the number of observations (n) must be greater than the number of independent variables (k), and that there is no perfect linear correlation among the independent variables.
📌 Example: A firm's sales (Y) are regressed on its advertising expenditures (X₁) and quality control expenditures (X₂) using 10 years of data. The estimated regression equation is: Ŷₜ = 17.944 + 1.873X₁ₜ + 1.915X₂ₜ The t-statistics for the coefficients are (2.663) for X₁ and (2.813) for X₂. With a critical t-value of 2.365 at the 5% significance level and 7 degrees of freedom, both parameters (coefficients) are statistically different from zero, meaning both advertising and quality control have a significant effect on sales.
ADJUSTED COEFFICIENT OF DETERMINATION
The adjusted R² (R̄²) modifies the R² to account for the number of independent variables in the model. As you add more variables, R² will always increase, even if the new variables are not truly significant. The adjusted R² penalizes the inclusion of unnecessary variables, providing a more realistic measure of the model's explanatory power, especially when comparing models with different numbers of variables. 📐 Formula: R̄² = 1 - (1 - R²) * ((n - 1) / (n - k)) → n is the number of observations, and k is the number of estimated parameters (including the intercept).
THE F STATISTIC
While R² tells us the proportion of variance explained, the F-statistic tests whether the independent variables as a group explain a statistically significant share of the variation in the dependent variable. It is the ratio of two variances. 📐 Formula: F = (Explained Variation / (k - 1)) / (Unexplained Variation / (n - k)) An equivalent formula using R² is: F = (R² / (k - 1)) / ((1 - R²) / (n - k)) → If the calculated F-statistic exceeds the critical value from the F-distribution table, we reject the hypothesis that all the regression coefficients (except the intercept) are zero, concluding that the overall model is statistically significant.
📌 Example: From the multiple regression example, R² = 0.930154, n=10, k=3. Calculated F = 46.61 Critical F at the 5% significance level = 4.74 Since 46.61 > 4.74, we conclude that the set of independent variables (advertising and quality control) as a group explains a statistically significant proportion of the total variation in sales.
PROBLEMS IN REGRESSION ANALYSIS
Several common problems can affect the validity of regression results:
- Multicollinearity: Two or more explanatory variables are highly correlated, making it difficult to isolate their individual effects. Standard errors of coefficients become large.
- Heteroskedasticity: The variance of the error term is not constant across all observations (e.g., it increases with Y). This violates a key regression assumption and can lead to inefficient estimates.
- Autocorrelation: Consecutive error terms are correlated. This is common in time series data and leads to incorrect standard errors.
🔑 Definition — Heteroskedasticity: The condition where the variance of the error term in a regression model is not constant across observations, often appearing as a funnel shape in a residual plot. 🔑 Definition — Autocorrelation: The correlation of a time series variable with its own past values. In regression, it refers to the correlation between error terms in different time periods.
DURBIN-WATSON STATISTIC
The Durbin-Watson (d) statistic is a formal test for autocorrelation. It is calculated from the residuals (errors) of the regression. 📐 Formula: d = ∑(eₜ - eₜ₋₁)² / ∑(eₜ)² → eₜ is the residual at time t, eₜ₋₁ is the residual at time t-1.
- d ranges from 0 to 4.
- If d = 2, there is no autocorrelation.
- If d is significantly less than 2, it suggests positive autocorrelation.
- If d is significantly greater than 2, it suggests negative autocorrelation.
⭐ Key Takeaways
The process of demand estimation using regression relies on several critical statistical tools and assumptions. The coefficient of determination (R²) is the primary measure to assess how well the independent variables explain the variation in demand, with a value closer to 1 indicating a better fit. For models with multiple variables, the adjusted R² provides a more accurate comparison by penalizing unnecessary terms. The F-statistic must be used to test whether the entire set of independent variables has a statistically significant explanatory power. Finally, it is essential to check for violations of regression assumptions, such as autocorrelation (using the Durbin-Watson statistic), multicollinearity, and heteroskedasticity, as these can invalidate the results and lead to poor business decisions.
🧠 Quick Revision Questions
- What are the five key assumptions of regression analysis regarding the error term?
- Explain the difference between the coefficient of determination (R²) and the adjusted coefficient of determination (R̄²). When is it more appropriate to use the adjusted one?
- In the context of the F-statistic, what is the null hypothesis being tested, and how is the result interpreted?
- What is the problem of multicollinearity in a multiple regression model, and what is one of its main consequences?
- What is the Durbin-Watson statistic used for, and what does a value of 2 indicate?