noc18-mg22 Lec 60-Linear Regression (Part-3)

IIT Kanpur July 201
20 followers
1 Video View·Dec 3, 2023
Hi, welcome back to the lecture series on microeconomics. We have been discussing basic
econometric theory and practice which is bread and butter for an applied microeconomist.
So, we have studied the ordinary least square estimators. First, we will finish the discussion
on the OLS estimators and then we will move to another related topic which is known as
hypothesis testing which is considered to be the heart and soul of econometrics.
Utilize a graph to show you what is exactly happening, then probably that will be more
effective. So, let me draw a 3D diagram where I measure my y or quantity demanded and my
price and the household income. So, this is price, okay. This is my income and y is quantity
demand, okay. So, now let us have the relationship plotted. So, let us first you know look at the
relationship between quantity demand and price. So, we expect a negatively sloped straight line,
right. So, let us first draw that, okay. So, this downward sloping straight line actually
has the slope value minus 087 and that is precisely the value of the partial regression
coefficient for a price variable, right, okay. And we expect a positively sloped straight line
for the x to y plane because as the income increases keeping the price fixed or constant,
then quantity demanded shall rise. So, we expect something like this and here the slope of this
red colored straight line is actually this estimated value of beta 2 parameter, beta 2 hat
equals to 11.1, okay. So, basically we are talking about a regression plane shaded by this area. So,
the next topic that we are going to study is you know to assess the goodness of fit of a regression.
So, basically the idea of regression analysis is to pass a straight line you know in between
scatter plot, right. So, if you have some data, a regression line should pass through the center
of the data and you know it should give a good fit to your data. So, by fit we mean that you
know it should be able to explain the variation in y well. So, how do we go about evaluation of
regression equation? So, we actually throw some explanatory variables in the regression model to
explain the variation in y. But you know if you think about the regression model,
then you will see that there is an intercept parameter beta 0 as well. So, what does that
mean? So, that means that if we do not throw any explanatory variable in the regression equation
to explain variation in y, then the variation will be explained by only a constant number and
you know an intercept term and that is basically the sample mean of the y. So,
the most naive explanatory variable is basically the constant term and that constant term is the
sample mean of the dependent variable y. So, after adding explanatory variables in the model,
whether I see some improvement in the explanatory power of my model to explain the variation in y.
Okay. So, now we have to look at some measures or concepts what we have already introduced and
that is basically the notion of total sum of squares, explained sum of squares and residual
sum of squares. So, basically let us first express the squared variations of y around its mean,
because that is the most naive explanatory variable. It is a measure of the amount of
variation to be explained by the sample mean only. So, let us first write down total sum of
square. Okay. Now, we have seen earlier that this can be broken down into two components and the
first component is called the explained sum of squares and the second component is known as the
residual sum of squares. Right. Okay. Then, the overall fit of the estimated regression model is
given by a concept called coefficient of determination and this coefficient of
determination is represented by symbol r square, notation r square and this is basically the ratio
of the explained sum of squares to the total sum of squares. This is not an econometrics course,
so I really do not want to get into deeper details of these concepts because that will
consume a lot of time, but there are many textbooks and online sources from where you
can study yourself. I am just you know going to give you the basic results what are required to
do some applied microeconomic data analysis. So, if you see some software output, then you
will understand what do these measures actually mean. Okay. So, now note that this r square takes
a value in a range. So, my TSS, RSS and ESS are all non-negative numbers, right because they are
the squared deviations and the explanatory sum of squares that is basically ESS is less than or
equal to TSS, the total sum of squares. So, by these two facts, I can safely write that my
coefficient of determination actually lies between 0 and 1. So, if my regression model
fails to analyze the variation in y altogether, then we will see the r square value will be 0
or very close to 0, it will be a very low fraction number, but if the regression model performs well
to explain the variations in y, it implies that we have chosen probably the perfect explanatory
variables to explain the variations in y, then we will observe a very high r square and at an
extreme if the explanatory variables are able to explain all variations in my dependent variable
y, then we will see r square value equal to 1, but that is just a theoretical possibility.
In practical purpose, you will probably not find r square value equal to 1. So,
now we are going to study what is known as classical linear regression model. So,
this model is basically not to be confused with classical economics or classical microeconomics,
classical here means basically the basic or simplest possible linear regression model.
So, let us start with a set of assumptions which are required to the formulation and
working of classical linear regression model. So, we will list down assumptions. So,
the first assumption is that the regression model is, okay. So, we have listed down all
the assumptions. Now, let us look at the explanation for some of the assumptions
in detail because they will help us in our analysis. So, assumption number 2,
assumption number 5 and assumption number 7 basically talks about a distribution for the
random error or stochastic noise in the regression model and we assume that this has a normal
distribution. So, I hope that you are aware of the normal distribution. So, I will not
get into details of normal distribution, okay. But what is the implication of these assumptions,
okay. So, the normality of error terms actually imply that these partial regression coefficients,
the betas, the estimated values of these betas which are basically beta hats are themselves
normally distributed because one of the property of the normal distribution says that any linear
function of normally distributed variables is itself normally distributed.
So, we can write down this implication or very interesting result that my regression
coefficients are normally distributed. One may ask why this is important to assume. Soon
we will find that we need to make some distributional assumptions for betas because that will help
us in testing economic theories, okay. So, we mean that my beta hat for any particular
explanatory variable has a distribution and the mean of the distribution for beta hat will be
beta, the population regression parameter, right and betas can take various values.
It can be say 0, it can be 1, it can be minus 2. So, all sort of possibilities are there. So,
what do we measure along the y axis in this diagram? So, just to complete the diagram,
let me have the vertical axis here. So, actually we measure probability, right.
So, now let us turn to the topic of hypothesis testing which is the heart and soul of econometric
theory. So, let me remind you about econometrics. What does an econometrician do actually? So,
an econometrician uses statistical techniques and economic theory to quantify the relationships
between economic variables using sample data, but you know does the journey of an econometrician
end right there after getting the estimates of population parameters beta 0, beta 1 and so on?
The answer is no. There is more to it. So, now once the regression equation has been established,
a pertinent question can be what do we learn from this regression equation? What do we conclude? So,
whatever fact we are observing from this sample regression equation, does it match with the
expectation from microeconomic theory or any other economic theory? So, that is to be done through
the lens of hypothesis testing which is originally borrowed from statistical methods.
So, to do hypothesis testing, we have to set proper hypothesis, right. So,
there could be two types of hypothesis that we use in practical hypothesis testing and
they are known as null hypothesis and alternative hypothesis. So, we will start
defining these two concepts. So, a null hypothesis is a statement on the range of
values of regression coefficient of interest that would be expected to be observed if the
theory were incorrect, right. So, basically the word null means zero as well. Null hypothesis
can also be thought as the hypothesis that the researcher does not believe in, ok.
There is also a concept of alternative hypothesis that we are going to study. So,
an alternative hypothesis uses, specifies the range of values of the regression coefficient
of interest or under consideration that would be expected to occur if the opposite happens,
it implies that the theory were indeed correct. So, hypothesis testing theory is a vast
area in statistics. We do not have time and also we do not require to know bit each and
every corner of this vast area, but I will just you know educate you on the basics of
hypothesis testing so that you can practice it in real life.
So, we will first start with the case of the demand function estimation problem and see
how hypothesis testing can be useful in order to test microeconomic theory. So, before we
jump on to the demand function example, so the notation used to refer a null hypothesis
is H0 and the alternative hypothesis is denoted usually by this notation, fine.
So, now let us look at an example of making null and alternative hypothesis using the
demand function estimation problem. Let us assume Y is the quantity demanded of a commodity
say car. So, a researcher can think about 3 or 4 variables which could explain the variation
in Y. The first one is of course the price of car, then X2 could be the income level
in the society and X3 could be the price of fuel, then X4 could be price of some alternative
commodity like two wheeler. So, there could be another one like the role of time because
tests and preferences change and economic conditions do change as well.
So, in a nutshell there could be a host of X factors which could explain variation in
Y but let me just work with only two of them as of now, okay. So, a simplified model could
be defined as Yi, i is basically the number of observations that we have, so that is basically
the size of the sample and I assume that it is a two variable regression model. Now note
that as X1 is price of car, my microeconomic theory says that there is law of demand which
actually dictates the relationship between X1 and Y and that is negative, right.
And we also know from the theory of consumer behavior that the relationship between Y and
X2 is expected to be positive, right. So, this is basically coming from my theory. So,
this is basically my hypothesis that I would like to test whether indeed the relationships
between these variables are following the theoretical result or not. So, we will continue
this discussion on hypothesis testing in the next lecture.
