Managerial Economics

Class 4

i. Clicker Review Questions

1. Statistical Significance
2. The i-Tunes Case using the Regression Tool
3. Regression Specification
a) Functional Form
b) Causation and Correlation
c) Multivariate Regression and Omitted Variables

Reading: 3.3, 3.4

THE UNIVERSITY OF BRITISH COLUMBIA
 Clicker Question 1
Which of the following statements about the Portland Cod regression is
false?

a. The demand curve we estimated was obtained by minimizing the sum

of squared residuals from the regression line.
b. Quantity is treated as the dependent quantity and price as the
explanatory variable.
c. Some observed points do not lie on the estimated demand curve.
d. Since the R-squared is quite high, this model is probably useful
e. None of the above.

Hint: See Regression Analysis Using Excel and Ordinary Least Squares
Regression, pp. 56-59.

THE UNIVERSITY OF BRITISH COLUMBIA

 Clicker Question 2
The Excel screenshot on the right estimates a demand
curve. Which statement is true?

a. The Excel Trendline only works for linear functions.

b. As set up, this Trendline will report the estimated
equation and R-squared on the graph.
c. In this case, the Trendline is set up to estimate a linear
equation but it could estimate a nonlinear form.
d. Ordinary least squares (OLS) makes the sum of residuals
as small as possible.
e. c. and d.

Hint: See Regression Analysis Using Excel and Ordinary

Least Square Regression, pp. 56-59.
THE UNIVERSITY OF BRITISH COLUMBIA
 1. Statistical Significance
When we gather some data we can always run a regression. But is the
result meaningful or just an accident or coincidence? If we repeat an
experiment many times and get a similar answer each time we are more
confident.

We cannot always repeat experiments, but we can use statistical techniques

to assess whether the results are significant.

We regard our results as meaningful if they have statistical significance.

Regression software (including Excel) calculates statistics to assess
statistical significance.

We cannot teach the underlying statistical theory here (but you will see it in
other courses). At this stage you just need to understand some terminology.
THE UNIVERSITY OF BRITISH COLUMBIA
 Terminology for Statistical
Significance
1. If we estimate a demand curve Q = a + bp, the estimated constants a and
b are estimated coefficients. (We expect b < 0.)

2. The standard error indicates how much the estimated coefficient would
vary in repeated samples. Small standard errors indicate precise
estimates.

3. A t-statistic equals the coefficient divided by the standard error. Larger

t-statistics indicate greater confidence that the true coefficient differs
from zero. (Being bigger than about 2 is important.)

4. A 95% confidence interval gives us bounds for an estimated coefficient

that will include the true value 95% of the time (assuming the model
specification is correct).
THE UNIVERSITY OF BRITISH COLUMBIA
 Install Analysis ToolPack

• For the next exercise, you will need to install analysis ToolPack on
your version of Excel

• For PC Excel: Click File  Options  Add-Ins

• In the Manage box, select Excel Add-Ins and click Go

• For Mac Excel: Click Tools  Excel Add-Ins

• In the Add-Ins box, check the Analysis ToolPack check box and
click OK

THE UNIVERSITY OF BRITISH COLUMBIA

 2. i-Tunes Regression in Excel using
focus group data
Quantity Price 1. Put this data in an Excel spreadsheet
with quantity in col. A and price in B.
441 1.49 2. Click Data  Data Analysis (on the top
493 1.29 right)  Regression
502 1.19 3. Input Y Range: Highlight the price data,
including the label.
536 1.09 4. Input X Range: Highlight the quantity
615 0.99 data, including the label.
643 0.89 5. Check the Labels and Confidence
Intervals boxes.
740 0.79 6. Choose the Output Range radio button,
757 0.69 and just select one cell beside your data
810 0.49 (e.g. E2 works well).

THE UNIVERSITY OF BRITISH COLUMBIA

 Clicker Question 3

The estimated i-Tunes demand curve is Q = 1024 – 413p.

a. At p = 0.99 the price elasticity is greater than 0.8 in absolute value.

b. At p = 0.99 the price elasticity is less than 0.8 in absolute value.
c. At p = 0.99 demand is elastic.
d. There is not enough information to determine the elasticity at p =0.99.
e. In this case we can calculate an arc elasticity but not a point elasticity.

THE UNIVERSITY OF BRITISH COLUMBIA

 t-statistics for the i-Tunes
Regression: Q = 1024 – 413p
The t-statistic for a regression coefficient is the estimated coefficient value
divided by the standard error.

For the price coefficient of -413 the estimated standard error is 32.7. The t-
statistic is therefore approximately -413/32.7 = -12.6.

This large t-statistic indicates that the coefficient is “statistically

significant” in the sense that we can be very confident it is different from
zero.

The 95% confidence interval for this coefficient is between -335 and -490.
(The calculation for this is not shown.)

We could say: “We would expect the true value to be between -335 and
-490 19 times out of 20.”
THE UNIVERSITY OF BRITISH COLUMBIA
 Clicker Question 4

In trying to assess whether an estimated demand curve is a useful guide to

real behaviour, which of the following is an indicator that our estimate is
useful?

a. A high R-squared value.

b. Small standard errors.
c. t-statistics that are large in absolute value.
d. Tight confidence intervals.
e. All of the above.

Hint: See “Confidence Intervals and Hypothesis Testing”, pp. 64-66.

THE UNIVERSITY OF BRITISH COLUMBIA

 iTunes Regression Results: Is
99¢ or 1.29 a better price?
The estimated regression line is Q = 1024 – 413p.

We can estimate the revenue, R, for any price:

R(p) = pQ(p) = p(1024 – 413p) = 1024p – 413 p2 .

If p = 0.99 then the estimated revenue is R = 609.

If p = 1.29 then the estimated revenue is R = 634.

Therefore, raising p from 0.99 to 1.29 would increase estimated revenue from
this focus group from 609 to 634.

THE UNIVERSITY OF BRITISH COLUMBIA

 Did it make sense for iTunes to
raise price?
We can use the
estimated demand
curve to calculate
revenue at different
prices, such as .99 and
1.24.

Revenue is pQ. In this

case $1.24 gives the
highest possible
revenue, more than at
99 cents (or $1.29).

THE UNIVERSITY OF BRITISH COLUMBIA

 Clicker Question 5
Assume that the cost of extra downloads to i-Tunes is zero. Therefore
maximizing profit is equivalent to maximizing revenue. Based on our focus
group analysis the profit-maximizing price for i-Tunes is therefore 1.24,
not 1.29. If we assume that Apple figured out the correct profit-maximizing
price, which of the following is a plausible reason for the difference?

a. Ordinary least squares is not a good statistical method in this case.

b. This focus group is not representative.
c. This focus group is representative but the resulting estimate is only an
approximation.
d. b. and c.
e. All of the above.

THE UNIVERSITY OF BRITISH COLUMBIA

 Revenue and Elasticity
We can use elasticity as a short-cut to tell us whether raising price will lower
or raise revenue, R.

Claim:
i) If demand is inelastic, then raising price will raise revenue.
ii) If demand is elastic, then raising price will lower revenue.
iii) If the elasticity = -1 then revenue is maximized and changing price
slightly in either direction will lower revenue.

Proof of i): (Using calculus)

R = p Q(p). Therefore, dR/dp = pdQ/dp + Q = Q(ε + 1).

If ε is less than 1 in absolute value (like -0.5) then dR/dp >0.
Therefore R rises if p rises.
THE UNIVERSITY OF BRITISH COLUMBIA
 Revenue and Elasticity cont’d

When price rises the price effect in itself tends to raise revenue, holding
quantity constant. But quantity falls, which tends to reduce revenue for
any given price. If demand is inelastic, then the percentage increase in
price exceeds the percentage decline in quantity so revenue rises.

If demand is elastic then the percentage decline in quantity is bigger

than the percentage increase in price, and revenue falls.

THE UNIVERSITY OF BRITISH COLUMBIA

 3. Regression Specification

In order to get good estimates from regression methods it is important

to have a good specification.

The specification relies on prior theory (possibly based on previous

analysis).

Among other things we need to specify an appropriate functional form,

and we need to identify the right explanatory variables.

THE UNIVERSITY OF BRITISH COLUMBIA

 Functional Form

Demand curves do not have to be linear and we do not have to estimate

linear functions. It is possible to estimate other functional forms. The
following figures show regressions of demand on advertising expenditure.
Is a linear function or a quadratic function a better fit for the data?
15.0
15.0

10.0 10.0

5.0 5.0

0.0
0.0
0 2 4 6 8 10 12 14 16 18
0 2 4 6 8 10 12 14 16 18

THE UNIVERSITY OF BRITISH COLUMBIA

 Causation and Correlation
Usually we use regression to estimate causal relationships – as when
price changes cause changes in quantity demanded. However,
regressions will produce high R-squared statistics and significant
coefficients even when a correlation is not causal.

Sunburn, S, and gasoline sales, G.

If we estimate G = a + bS using monthly data from Canada we
would find a significant coefficient on S and a high R-squared.

Umbrellas, U, and rain, R.

If we estimate the regression R = a + bU on weekly data for stores in
Vancouver we find a significant coefficient and a high R-squared.

Are these good regressions?

THE UNIVERSITY OF BRITISH COLUMBIA
Causation and Correlation cont’d

Suppose we regress Y on X: Y = a + bX.

A correlation between Y and X might arise if

X causes Y.
Y causes X (reverse causality).
X and Y are both caused by some third factor
X and Y are unrelated but both have a strong time
trend.

Other examples:
artificial sweeteners and BMI
police and crime rates
lemon imports and traffic accidents
THE UNIVERSITY OF BRITISH COLUMBIA
Cause and Effect?

THE UNIVERSITY OF BRITISH COLUMBIA

 Clicker Question 6
If we regress crime rates on police presence using current data from
U.S. cities interpretation of the result is difficult because of

a. A third factor problem

b. Reverse causality
c. Causality running in both directions
d. Time trends
e. All of the above

THE UNIVERSITY OF BRITISH COLUMBIA

 Multivariate Regression
If we estimate a demand curve of the form Q = a + bp, then Q is the dependent
variable and there is only one explanatory variable (price, p). Therefore we
call this a univariate regression.

Estimating such a regression equation makes sense if other important

variables, such as income, are held constant.

However, if both price and income vary in our data, we might wish to estimate
a demand function with two explanatory variables – price and income:
Q = a + bp + cY, where Y represents income.

This is a multivariate regression – there is more than one explanatory variable.

See the mini-case on CEO compensation (p. 67) for a multivariate example.
THE UNIVERSITY OF BRITISH COLUMBIA
 Summary of Regression
1. We obtain estimates of demand curves (and other relationships)
using regression.
2. A univariate regression is like drawing a line through some points.
3. We can use Excel to run regressions, with the Scatterplot Trendline
function and the more sophisticated Regression tool.
4. There are various measures that help us interpret the significance
of our results.
5. It is important to have an appropriate regression specification
(functional form and included variables).

THE UNIVERSITY OF BRITISH COLUMBIA