Session 3

• Quantitative Demand
Analysis – Estimation of
Demand
 Regression Analysis
• Used to estimate demand functions
• Important terminology
– Least Squares Regression: Y = a + bX + e
– Confidence Intervals
– t-statistic
– R-square or Coefficient of Determination
– F-statistic
 Regression Analysis
 Statistical technique for estimating a
relationship between the dependent
variable and one or more
independent variables
 Ex: relationship between advertising
expenditures and sales revenues.
Regression Analysis
 If we plot data for x and y on a
graph--
 Regression analysis estimates the
line that minimizes the sum of the
squared vertical deviations of the
observations from the line.
 Regression Analysis -
Scatter Diagram
Y
.
. Y = a + bX
.
. .
. .
X
Objective Of Regression Analysis

 To obtain estimates of A & B

 B (SLOPE COEFFICIENT)
measures increase in dependent
variable resulting from unit
increase in independent variable
 a (INTERCEPT) gives value of Y
when X = 0.
 Simple Linear Regression OLS --
• Qt = a + b Pt + t Q ordinary
least
• time subscripts & error squares
term
• Find “best fitting” line _
t = Qt - a - b Pt Q

t 2= [Qt - a - b Pt] 2
• mint 2= [Qt - a - b Pt] 2
• Solution: b = _
Cov(Q,P)/Var(P) and a = P
mean(Q) - b•mean(P)
 Ordinary Least Squares
 e1 is vertical deviation or error of actual
Y from Y estimated from regression line
in first year:
 e1 = Y1 - Y’1
 errors occur because:
 many variables have slight affect on Y
 error measurement in Y
 random human error
 Ordinary Least Squares
 Computer programs can be used to
estimate regression line
 For example
 Yt = 7.60 + 3.53Xt
 By substituting a value for Xt, the
regression line can estimate Yt
 the b measures the marginal effect on Y
from each unit change in X
 Ordinary Least Squares:
Assumptions & Solution Methods

• error term has a • Spreadsheets -

mean of zero and a • Statistical calculators
finite variance Minitab, SAS, SPSS
• dependent variable • ForeProfit
is random • Excel, Lotus, Quatro Pro,
• the independent Joe Spreadsheet
variables are indeed 
tools/data analysis in Excel
independent 
/Data/Regression in Lotus
 Using a Spreadsheet to
Perform a Regression
• Pages 94-95
• Homework Questions 22, 23 and 24
are regression analysis problems
using Excel
• Spreadsheet program produces
detailed information about
regression
 Tests Of Significance
 to test whether independent
variable positively affects the
dependent variable
 To test for the statistical
significance of , we use standard
error (deviation) of  given by Sb
 t-statistic = /S we get the t-statistic
 Tests Of Significance

 Rule of thumb - if the value of the t-

statistic is greater than or equal to 2,
then the parameter estimate is
statistically different from zero.
 the higher the calculated t-ratio, the
more confident we are that the true
value of  is not equal to zero
 P-Values
• Regression packages report P-values
• much more precise measure of statistical
significance
• If P-value for an estimated coefficient is .0009,
this means there is only a 9 in 10,000 chance that
the true coefficient of the variable is actually
zero.
• P-values of .05 or below are considered
significant.
 Confidence Intervals
• Firm manager can construct upper
and lower bounds on the true value
of the estimated coefficient by
constructing a 95% confidence
interval.
• Rule of thumb for coefficients (a
and b) is:
 a or - 2 a
 b + or - 2 
Coefficient Of Determination
(R2)
 Measures proportion of total
variation in Y explained by
variation in X.
 If R2 = 78%, then variation in X
explains 78% of variation in Y.
(Ie: 22% of variation is explained
by something else.)
Evaluates the overall fit of the
regression line
 This tests for the overall
explanatory power of the entire
regression.
 If R2 = 1, all the variation in the
dependent variable would be
explained by the variation in
independent variables.
 Test Of Goodness Of Fit And
Correlation

 The square root of R2 is the

coefficient of correlation (r).
 r is the measure of the degree
of covariation that exists
between x and y
 r ranges from -1 to +1.
 Adjusted R 2

• Problem with R2 is that as we add more

coefficients, the R2 increases.
• We may have a high R2 because the
number of observations is small relative
to the number of estimated parameters.
• provides a misleading indicator of
goodness of fit
• Better measure is adjusted R-square

R2 = 1 - (1 - R2)[(n-1)/n-k)]
 The F-Statistic
• An alternative measure of regression
line “good fit” is F-Statistic
• F-Statistic provides measure of total
variation explained by the regression
relative to the total unexplained
variation.
• the greater the F-statistic, the better
the overall fit of the regression line.
 Nonlinear Regression
• For example, if the relationship is a
curve.
• For a log-linear demand function,
you would take the logarithm of
prices and quantities before
executing the regression.
• You would regress the transformed
Q on P to obtain parameter
estimates.
 An Example

• Use a spreadsheet to estimate

log-linear demand

log Qx   0   x log Px  e
 Summary Output

Regression Statistics
Multiple R 0.41
R Square 0.17
Adjusted R Square 0.15
Standard Error 0.68
Observations 41.00

ANOVA
df SS MS F Significance F
Regression 1.00 3.65 3.65 7.85 0.01
Residual 39.00 18.13 0.46
Total 40.00 21.78

Coefficients Standard Error t Stat P-value Lower 95% Upper 95%

Intercept 7.58 1.43 5.29 0.000005 4.68 10.48
ln(P) -0.84 0.30 -2.80 0.007868 -1.44 -0.23
 Interpreting the Output

• Estimated demand function:

– log Qx = 7.58 - 0.84 logPx
– Own price elasticity: -0.84 (inelastic)
• How good is our estimate?
– t-statistics of 5.29 and -2.80 indicate that the
estimated coefficients are statistically different from
zero
– R-square of .17 indicates we explained only 17
percent of the variation
– F-statistic significant at the 1 percent level.
Regression Analysis With
Excel
Page 3-7 in packet of
handouts
 Regression Example
Assume that bus ridership is a function of
the price of the ticket, population density,
per capita income levels in an area, and
the number of parking places available.
Given data, you could conduct a
regression analysis to try to estimate
demand for bus ridership.
The results may look like those in the
following slide.
 Demand Estimation Case
Riders = 785 -2.14•Price +.110•Pop
+.0015•Income + .995•Parking
Predictor Coef St.dev t-ratio p
Constant 784.7 396.3 1.98 .083
Price -2.14 .4890 -4.38 .002
Pop .1096 .2114 .520 .618
Income .0015 .03534 .040 .966
Parking .9947 .5715 1.74 .120
R-sq = 90.8% R-sq(adj) = 86.2%
Regression example
A public agency responsible for serving the
commuter rail transportation needs of a large
city is facing rising operating deficits on its
system.
The board of directors asked the system
manger to explore alternatives to alleviate the
financial plight of the system
One suggestion is to institute a major cutback
in service
The board suggested that the system manager
look into increasing fares and the conduct a
study of the likely impact of a proposed fare
hike.
 Regression example
The system manager has collected data
on important variables thought to have
an impact on the demand for rides:
Price per ride (in cents) = P = -
Population in the area serviced = T = +
Disposable per capita income = I = +
Parking rate per hour in the downtown area
(in cents) = H = +
Perform a multiple regression on the
data to determine the impact of the rate
increase. Data follows on next slide.
 Regression Data
Year Weekly Price Populati Income Parking
Riders (cents) on (T) (I) Rate
(H)
1984 1,200 15 1800 2900 50
1985 1,190 15 1790 3100 50
1986 1,195 15 1780 3200 60
1987 1,110 25 1,778 3,250 60
1988 1,105 25 1,750 3,275 60
1989 1,115 25 1740 3290 70
1990 1,130 25 1725 4100 75
1991 1,095 30 1725 4300 75
1992 1,090 30 1720 4400 75
Demonstration Problem 3-5
• Page 101
• Multiple Regression
 SUMMARY:
Given market or survey data, regression
analysis can be used to estimate:

Demand functions
Elasticities
A host of other things, including cost
functions
Managers can quantify the impact of changes
in prices, income, advertising, etc.