Lecture 9

Effect of the change in scale: dependent variable

• Consider the regression:
• 𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + ⋯ . . +𝛽𝑘 𝑥𝑘 + 𝑢
• Suppose instead of this we run the following regression:
• 𝑐𝑦 = 𝛾0 + 𝛾1 𝑥1 + 𝛾2 𝑥2 + ⋯ . . +𝛾𝑘 𝑥𝑘 + 𝑢
• The OLS estimates of 𝛾0 , 𝛾1 , … . 𝛾𝑘 will be scaled c times.
• 𝛾ො0 = 𝑐𝛽መ0 , 𝛾ො1 = 𝑐𝛽መ1 , 𝛾ො2 = 𝑐 𝛽መ2 , … … . . , 𝛾ො𝑘 = 𝑐𝛽መ𝑘
Effect of the change in scale: independent variable
• Consider the regression:
• 𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + ⋯ . . +𝛽𝑘 𝑥𝑘 + 𝑢
• Suppose instead of this we run the following regression:
• 𝑦 = 𝛾0 + 𝛾1 (𝑐1 𝑥1 ) + 𝛾2 𝑥2 + ⋯ . . +𝛾𝑘 𝑥𝑘 + 𝑢
• The OLS estimate 𝛾1 will be divided by c. Other estimates remains unchanged
• 𝛾ො0 = 𝛽መ0 , 𝛾ො1 = 𝛽መ1 /𝑐, 𝛾ො2 = 𝛽መ2 , … … . . , 𝛾ො𝑘 = 𝛽መ𝑘
• The OLS estimate 𝛾1 will be divided by c. Other estimates remains unchanged
Question
• Can we include both 𝑥1 and 𝑐1 𝑥1 as regressors in the same
regression?
 Example: Effect of change of scaling
• Here dependent variables:
• Birthweight: measured in two units
• bwght: ounces
• bwghtlbs: pounds
• bwghtlbs= bwght/16
• Independent variables
• Smoking by parents: two units
• Number of packets: packs
• Number of cigarettes:cigs
• packs=cigs/20
• Family income
• Each column represents a separate regression:
dependent variables mentioned as column
headers
• The independent variables are mentioned in
the rows
• Note that cigs and packs are not included in the
same regression since packs is a multiple of cigs.
Changing the scale of dependent when the
dependent variable in in log form
• Consider the regression:
• ln 𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + ⋯ . . +𝛽𝑘 𝑥𝑘 + 𝑢
• Suppose instead of this we run the following regression:
• ln(𝑐1 𝑦) = 𝛾0 + 𝛾1 𝑥1 + 𝛾2 𝑥2 + ⋯ . . +𝛾𝑘 𝑥𝑘 + 𝑢
• In the simple regression model with one variable, changing the unit of
measurement does not affect the slope coefficient.
• The same is true here: changing the unit of measurement of the dependent variable,
when it appears in logarithmic form, does not affect any of the slope estimates.
• Only theintercept changes
• This follows from the simple fact that ln 𝑐1 𝑦 = ln 𝑐1 + ln 𝑦 for any
constant 𝑐1 > 0.
Changing the scale of independent variable
when the dependent variable in in log form
• Consider the regression:
• 𝑦 = 𝛽0 + 𝛽1 ln 𝑥1 + 𝛽2 𝑥2 + ⋯ . . +𝛽𝑘 𝑥𝑘 + 𝑢
• Suppose instead of this we run the following regression:
• 𝑦 = 𝛾0 + 𝛾1 ln 𝑐1 𝑥1 + 𝛾2 𝑥2 + ⋯ . . +𝛾𝑘 𝑥𝑘 + 𝑢
• No change in any slope estimates
• Changing the unit of measurement of any 𝑥𝑗 , where ln 𝑥𝑗 appears in
the regression, only affects the intercept.
Model with Quadratics
• Quadratic functions are also used quite often in applied economics to
capture decreasing or increasing marginal effects.
In the simplest case, y depends on a single observed factor x, but it does so
in a quadratic fashion:
• 𝑦 = 𝛽0 + 𝛽1 𝑥 + 𝛽2 𝑥 2 + 𝑢
• It is important to remember that 𝛽1 does2 not measure the change in y with
respect to x; it makes no sense to hold 𝑥 fixed while changing x.
• If we write the estimated equation as
ෝ = 𝛽መ0 + 𝛽መ1 𝑥 + 𝛽መ2 𝑥 2
𝒚
• Then we have:
𝜕ෝ𝒚
• = 𝛽መ1 + 2𝛽መ2 𝑥
𝜕𝑥
• If we plug in 𝑥 = 0, we see that 𝛽መ1 can be interpreted as the approximate
slope in going from x=0 to x=1.
• After that, the second term, 2𝛽መ2 𝑥, must be accounted for.
• The marginal increase in y when x changes from 1 to 2 is 𝛽መ1 + 2𝛽መ2 .
 Example1: quadratics in regression

• Marginal effect of exper:

• 0.298-2*0.0061*exper
• This estimated equation implies that
exper has a diminishing effect on wage:
• Inverted u-shaped curve
• The tip of the inverted U occurs when
the slope is zero, or
• 𝑥 = −𝛽መ1 /2𝛽መ2
• Thus in the wage equation, the return to
experience becomes zero at about 24.4
years.
Example 2: quadratics in regression

• Thus

• In other words
Models with interactions
• Sometimes, it is natural for the partial effect with respect to an explanatory variable to
depend on the magnitude of yet another explanatory variable.
• For example, in the model

• In this case, the partial effect of 𝑏𝑑𝑟𝑚𝑠 𝑜𝑛 𝑝𝑟𝑖𝑐𝑒:

෣
𝜕𝑝𝑟𝑖𝑐𝑒
• = 𝛽መ2 + 𝛽መ3 𝑠𝑞𝑟𝑓𝑡
𝜕𝑏𝑑𝑟𝑚𝑠
• If 𝛽መ3 > 0, an additional bedroom yields a higher increase in housing price for houses
with higher sqrft.
• In other words, there is an interaction effect between square footage and number of
bedrooms.
• The parameters on the original variables can be tricky to interpret when we include an
interaction term. For example, 𝛽2 is the effect of bdrms on price for a home with zero
square feet.
• This effect is clearly not of much interest. Instead, we must be careful to put interesting values of
sqrft, such as the mean or median values in the sample, into the estimated version of equation to
interpret the effect of bdrms on price
Adjusted R-Squared
• Most regression packages will report, along with the R-squared, a
statistic called the adjusted R-squared. Because the adjusted R-
squared is reported in much applied work, and because it has some
useful features, we cover it in this subsection.
• Note that R-squared can be written as:

• The adjusted R-squared is defined as:

• The primary attractiveness of adjusted-𝑅2 is that it imposes a penalty
for adding additional independent variables to a model.
• We know that 𝑅2 can never fall when a new independent variable is
added to a regression equation: this is because SSR never goes up
(and usually falls) as more independent variables are added.
• But the formula for adjusted-𝑅2 shows that it depends explicitly on k, the
number of independent variables.
• If an independent variable is added to a regression, SSR falls, but so
does the df in the regression, 𝑛 − 𝑘 − 1.
𝑆𝑆𝑅
• can go up or down when a new independent variable is added
𝑛−𝑘−1
to a regression.
Controlling for too many factors in the
regressions
• To avoid the mistake of controlling too many factors, we need to
remember the ceteris paribus interpretation of multiple regression
models.
• To illustrate this issue, suppose we are doing a study to assess the
impact of state beer taxes on traffic fatalities.
• The idea is that a higher tax on beer will reduce alcohol consumption, and
likewise drunk driving, resulting in fewer traffic fatalities.
• To measure the ceteris paribus effect of taxes on fatalities, we can model
fatalities as a function of several factors, including the beer tax
•
• Notice how we have not included a variable measuring per capita
beer consumption. Are we committing an omitted variables error?
The answer is no. If we control for beer consumption in this equation,
then how would beer taxes affect traffic fatalities? In the equation

• 𝛽1 measures the difference in fatalities due to a one percentage point

increase in tax, holding beercons fixed.
• It is difficult to understand why this would be interesting. We should
not be controlling for differences in beercons across states, unless we
want to test for some sort of indirect effect of beer taxes.
• The previous examples are what can be called over controlling for
factors in multiple regression.
• If we focus on the ceteris paribus interpretation of regression, then
we will not include the wrong factors in a regression