Você está na página 1de 84

Pooling Cross-Sections Across Time

Modelling Time Eects in Cross-Sections


Chow Tests of Structural Change
Panel Data
Summary

Cross-Secctions and Panels with Two Periods


Quantitative Microeconomics

R. Mora

Department of Economics
Universidad Carlos III de Madrid

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Outline

Pooling Cross-Sections Across Time

Modelling Time Eects in Cross-Sections

Chow Tests of Structural Change

Panel Data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

What is Panel Data?

up to now, we have seen only cross-sections

each observation represents an individual, rm, etc


there are also time series: each observation represents a
separate period
we may also have cross sections for dierent time periods
(pooled cross-sections)
and we may also have panel data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

What is Panel Data?

up to now, we have seen only cross-sections

each observation represents an individual, rm, etc


there are also time series: each observation represents a
separate period
we may also have cross sections for dierent time periods
(pooled cross-sections)
and we may also have panel data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

What is Panel Data?

up to now, we have seen only cross-sections

each observation represents an individual, rm, etc


there are also time series: each observation represents a
separate period
we may also have cross sections for dierent time periods
(pooled cross-sections)
and we may also have panel data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

What is Panel Data?

up to now, we have seen only cross-sections

each observation represents an individual, rm, etc


there are also time series: each observation represents a
separate period
we may also have cross sections for dierent time periods
(pooled cross-sections)
and we may also have panel data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

What is Panel Data?

up to now, we have seen only cross-sections

each observation represents an individual, rm, etc


there are also time series: each observation represents a
separate period
we may also have cross sections for dierent time periods
(pooled cross-sections)
and we may also have panel data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

What is Panel Data?

up to now, we have seen only cross-sections

each observation represents an individual, rm, etc


there are also time series: each observation represents a
separate period
we may also have cross sections for dierent time periods
(pooled cross-sections)
and we may also have panel data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Pooled Cross Sections


CIS surveys
each month independently, the Centro de Investigaciones
Sociolgicas samples the Spanish population
the surveys ask for political and sociological actitudes
individuals interviewed in one month are not interviewed in any
other month
data with this structure are known as pooled cross-sections
they are cross-sections, but you may study time eects

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Pooled Cross Sections


CIS surveys
each month independently, the Centro de Investigaciones
Sociolgicas samples the Spanish population
the surveys ask for political and sociological actitudes
individuals interviewed in one month are not interviewed in any
other month
data with this structure are known as pooled cross-sections
they are cross-sections, but you may study time eects

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Pooled Cross Sections


CIS surveys
each month independently, the Centro de Investigaciones
Sociolgicas samples the Spanish population
the surveys ask for political and sociological actitudes
individuals interviewed in one month are not interviewed in any
other month
data with this structure are known as pooled cross-sections
they are cross-sections, but you may study time eects

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Pooled Cross Sections


CIS surveys
each month independently, the Centro de Investigaciones
Sociolgicas samples the Spanish population
the surveys ask for political and sociological actitudes
individuals interviewed in one month are not interviewed in any
other month
data with this structure are known as pooled cross-sections
they are cross-sections, but you may study time eects

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Panel Data


ECHP 19942001
each year, the European Community Household Panel surveys
the same individuals on a wide range of topics: income, health,
education, housing, employment, etc
in the rst wave, i.e. in 1994, approximately 130,000 adults
aged 16 years and over

in panel data, we have observations for each individual for at


least two periods
panel data is also referred to as longitudinal data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Panel Data


ECHP 19942001
each year, the European Community Household Panel surveys
the same individuals on a wide range of topics: income, health,
education, housing, employment, etc
in the rst wave, i.e. in 1994, approximately 130,000 adults
aged 16 years and over

in panel data, we have observations for each individual for at


least two periods
panel data is also referred to as longitudinal data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Panel Data


ECHP 19942001
each year, the European Community Household Panel surveys
the same individuals on a wide range of topics: income, health,
education, housing, employment, etc
in the rst wave, i.e. in 1994, approximately 130,000 adults
aged 16 years and over

in panel data, we have observations for each individual for at


least two periods
panel data is also referred to as longitudinal data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

An Example of Panel Data


ECHP 19942001
each year, the European Community Household Panel surveys
the same individuals on a wide range of topics: income, health,
education, housing, employment, etc
in the rst wave, i.e. in 1994, approximately 130,000 adults
aged 16 years and over

in panel data, we have observations for each individual for at


least two periods
panel data is also referred to as longitudinal data

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Why using Pooled Cross Sections


we may want to pool cross sections just to get bigger samples
we may want to capture time eects
suppose we assume that parameters remain constant
wages

it = 0 + 1 educ it + uit

alternatively, we may want to investigate the eect of time on


the parameters of the model
to do so, we need to make assumptions about the value of the
parameters in each period

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Why using Pooled Cross Sections


we may want to pool cross sections just to get bigger samples
we may want to capture time eects
suppose we assume that parameters remain constant
wages

it = 0 + 1 educ it + uit

alternatively, we may want to investigate the eect of time on


the parameters of the model
to do so, we need to make assumptions about the value of the
parameters in each period

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Why using Pooled Cross Sections


we may want to pool cross sections just to get bigger samples
we may want to capture time eects
suppose we assume that parameters remain constant
wages

it = 0 + 1 educ it + uit

alternatively, we may want to investigate the eect of time on


the parameters of the model
to do so, we need to make assumptions about the value of the
parameters in each period

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Why using Pooled Cross Sections


we may want to pool cross sections just to get bigger samples
we may want to capture time eects
suppose we assume that parameters remain constant
wages

it = 0 + 1 educ it + uit

alternatively, we may want to investigate the eect of time on


the parameters of the model
to do so, we need to make assumptions about the value of the
parameters in each period

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Why using Pooled Cross Sections


we may want to pool cross sections just to get bigger samples
we may want to capture time eects
suppose we assume that parameters remain constant
wages

it = 0 + 1 educ it + uit

alternatively, we may want to investigate the eect of time on


the parameters of the model
to do so, we need to make assumptions about the value of the
parameters in each period

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Simplest Model with Time Eects


the simplest model assumes that only the intercept changes
wages

it = 0t + 1 educ it + uit

where t

= 1, ..., T

as the constant only captures scale eects, this model allows


for changes in the units of measurement of the dependent
variable across periods
this eectively controls for annual ination in a wage
regression model if wages are nominal wages
in general, it captures time changes in the denition of the
dependent variable which aect all observations equally

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Simplest Model with Time Eects


the simplest model assumes that only the intercept changes
wages

it = 0t + 1 educ it + uit

where t

= 1, ..., T

as the constant only captures scale eects, this model allows


for changes in the units of measurement of the dependent
variable across periods
this eectively controls for annual ination in a wage
regression model if wages are nominal wages
in general, it captures time changes in the denition of the
dependent variable which aect all observations equally

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Simplest Model with Time Eects


the simplest model assumes that only the intercept changes
wages

it = 0t + 1 educ it + uit

where t

= 1, ..., T

as the constant only captures scale eects, this model allows


for changes in the units of measurement of the dependent
variable across periods
this eectively controls for annual ination in a wage
regression model if wages are nominal wages
in general, it captures time changes in the denition of the
dependent variable which aect all observations equally

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Simplest Model with Time Eects


the simplest model assumes that only the intercept changes
wages

it = 0t + 1 educ it + uit

where t

= 1, ..., T

as the constant only captures scale eects, this model allows


for changes in the units of measurement of the dependent
variable across periods
this eectively controls for annual ination in a wage
regression model if wages are nominal wages
in general, it captures time changes in the denition of the
dependent variable which aect all observations equally

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (1/2)


we can investigate whether relations change in time
wages

it = 0 + 1t educ it + uit

where t=1,...,T

returns to education may change every period


we have as many slopes for education as time periods
an alternative presentation of the same model
wages

it = 0 + 11 educit 1(t = 1) + ...+


1T educit 1(t = T ) + uit

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (1/2)


we can investigate whether relations change in time
wages

it = 0 + 1t educ it + uit

where t=1,...,T

returns to education may change every period


we have as many slopes for education as time periods
an alternative presentation of the same model
wages

it = 0 + 11 educit 1(t = 1) + ...+


1T educit 1(t = T ) + uit

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (1/2)


we can investigate whether relations change in time
wages

it = 0 + 1t educ it + uit

where t=1,...,T

returns to education may change every period


we have as many slopes for education as time periods
an alternative presentation of the same model
wages

it = 0 + 11 educit 1(t = 1) + ...+


1T educit 1(t = T ) + uit

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (1/2)


we can investigate whether relations change in time
wages

it = 0 + 1t educ it + uit

where t=1,...,T

returns to education may change every period


we have as many slopes for education as time periods
an alternative presentation of the same model
wages

it = 0 + 11 educit 1(t = 1) + ...+


1T educit 1(t = T ) + uit

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Time Eects in the Returns to Education (2/2)


for simplicity, assume two periods (T
wages

= 2)

it = 0 + 11 educit 1(t = 1)+


12 educit 1(t = 2) + uit

This expression suggests a simple way to estimate the model


Step 1: generate T interactions with educ

=
educ 2 =

genr educ 1
genr

(year == 1)
educ (year == 2)
educ

Step 2: ols wages const educ 1 educ 2


with T

= 3,

it

robust

we need three interactions, with T

= 4,

four, ...

each slope measures the returns to education each year

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (1/2)


wages

it = 0 + 11 educit 1(t = 1) + 12 educit 1(t = 2) + uit

frequently, what we want to see is whether there has been any


change in the returns to education
H0

: 12 = 11 H0 :12 11 = 0

for a sample with two periods, 1(t

= 1) = 1 1(t = 2)

hence, the model can also be expressed as:

wages

it = 0 + 11 educit + (12 11 ) educit 1(t = 2) + uit


R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (1/2)


wages

it = 0 + 11 educit 1(t = 1) + 12 educit 1(t = 2) + uit

frequently, what we want to see is whether there has been any


change in the returns to education
H0

: 12 = 11 H0 :12 11 = 0

for a sample with two periods, 1(t

= 1) = 1 1(t = 2)

hence, the model can also be expressed as:

wages

it = 0 + 11 educit + (12 11 ) educit 1(t = 2) + uit


R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (1/2)


wages

it = 0 + 11 educit 1(t = 1) + 12 educit 1(t = 2) + uit

frequently, what we want to see is whether there has been any


change in the returns to education
H0

: 12 = 11 H0 :12 11 = 0

for a sample with two periods, 1(t

= 1) = 1 1(t = 2)

hence, the model can also be expressed as:

wages

it = 0 + 11 educit + (12 11 ) educit 1(t = 2) + uit


R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (1/2)


wages

it = 0 + 11 educit 1(t = 1) + 12 educit 1(t = 2) + uit

frequently, what we want to see is whether there has been any


change in the returns to education
H0

: 12 = 11 H0 :12 11 = 0

for a sample with two periods, 1(t

= 1) = 1 1(t = 2)

hence, the model can also be expressed as:

wages

it = 0 + 11 educit + (12 11 ) educit 1(t = 2) + uit


R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (1/2)


wages

it = 0 + 11 educit 1(t = 1) + 12 educit 1(t = 2) + uit

frequently, what we want to see is whether there has been any


change in the returns to education
H0

: 12 = 11 H0 :12 11 = 0

for a sample with two periods, 1(t

= 1) = 1 1(t = 2)

hence, the model can also be expressed as:

wages

it = 0 + 11 educit + (12 11 ) educit 1(t = 2) + uit


R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (1/2)


wages

it = 0 + 11 educit 1(t = 1) + 12 educit 1(t = 2) + uit

frequently, what we want to see is whether there has been any


change in the returns to education
H0

: 12 = 11 H0 :12 11 = 0

for a sample with two periods, 1(t

= 1) = 1 1(t = 2)

hence, the model can also be expressed as:

wages

it = 0 + 11 educit + (12 11 ) educit 1(t = 2) + uit


R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (2/2)


H0

Step 1: generate T

:12 11 = 0

interactions

genr educ 2=educ

(year == 2)

Step 2
ols wages const educ educ 2

robust

the slope of the interaction(s) measures how that year's slope


diers from the slope of the rst year
we can choose the reference year at will

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (2/2)


H0

Step 1: generate T

:12 11 = 0

interactions

genr educ 2=educ

(year == 2)

Step 2
ols wages const educ educ 2

robust

the slope of the interaction(s) measures how that year's slope


diers from the slope of the rst year
we can choose the reference year at will

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (2/2)


H0

Step 1: generate T

:12 11 = 0

interactions

genr educ 2=educ

(year == 2)

Step 2
ols wages const educ educ 2

robust

the slope of the interaction(s) measures how that year's slope


diers from the slope of the rst year
we can choose the reference year at will

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (2/2)


H0

Step 1: generate T

:12 11 = 0

interactions

genr educ 2=educ

(year == 2)

Step 2
ols wages const educ educ 2

robust

the slope of the interaction(s) measures how that year's slope


diers from the slope of the rst year
we can choose the reference year at will

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (2/2)


H0

Step 1: generate T

:12 11 = 0

interactions

genr educ 2=educ

(year == 2)

Step 2
ols wages const educ educ 2

robust

the slope of the interaction(s) measures how that year's slope


diers from the slope of the rst year
we can choose the reference year at will

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Testing Time Changes in the Slopes (2/2)


H0

Step 1: generate T

:12 11 = 0

interactions

genr educ 2=educ

(year == 2)

Step 2
ols wages const educ educ 2

robust

the slope of the interaction(s) measures how that year's slope


diers from the slope of the rst year
we can choose the reference year at will

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Independent Regressions
What happens if we allow for parameter time variation in all years?

wages

it = 0t + 1t educ it + uit

all beta coecients estimated with a sample of N observations


the standard deviation estimated with a sample of NT
observations
for the slopes, it is exactly like estimating all periods separately
for inference, it is dierent if we assume a constant error
variance in the pooled estimation

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Independent Regressions
What happens if we allow for parameter time variation in all years?

wages

it = 0t + 1t educ it + uit

all beta coecients estimated with a sample of N observations


the standard deviation estimated with a sample of NT
observations
for the slopes, it is exactly like estimating all periods separately
for inference, it is dierent if we assume a constant error
variance in the pooled estimation

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Independent Regressions
What happens if we allow for parameter time variation in all years?

wages

it = 0t + 1t educ it + uit

all beta coecients estimated with a sample of N observations


the standard deviation estimated with a sample of NT
observations
for the slopes, it is exactly like estimating all periods separately
for inference, it is dierent if we assume a constant error
variance in the pooled estimation

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Independent Regressions
What happens if we allow for parameter time variation in all years?

wages

it = 0t + 1t educ it + uit

all beta coecients estimated with a sample of N observations


the standard deviation estimated with a sample of NT
observations
for the slopes, it is exactly like estimating all periods separately
for inference, it is dierent if we assume a constant error
variance in the pooled estimation

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Independent Regressions
What happens if we allow for parameter time variation in all years?

wages

it = 0t + 1t educ it + uit

all beta coecients estimated with a sample of N observations


the standard deviation estimated with a sample of NT
observations
for the slopes, it is exactly like estimating all periods separately
for inference, it is dierent if we assume a constant error
variance in the pooled estimation

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Independent Regressions
What happens if we allow for parameter time variation in all years?

wages

it = 0t + 1t educ it + uit

all beta coecients estimated with a sample of N observations


the standard deviation estimated with a sample of NT
observations
for the slopes, it is exactly like estimating all periods separately
for inference, it is dierent if we assume a constant error
variance in the pooled estimation

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

The Chow Test for Structural Change


wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 01 = 02 , 11 = 12

compute the F test

the test is rejected if there is a signicant change in any of the


slopes or in the constant
changes in the constant sometimes only capture dierences in
the measurement of the dependent variable
in these cases it seems more interesting to test only structural
change in the slopes

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Other Versions of the Test

wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 11 = 12

compute the F test

it is possible to test change in the slopes of several variables


simultaneously

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Other Versions of the Test

wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 11 = 12

compute the F test

it is possible to test change in the slopes of several variables


simultaneously

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Other Versions of the Test

wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 11 = 12

compute the F test

it is possible to test change in the slopes of several variables


simultaneously

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Other Versions of the Test

wages

it = 0t + 1t educit + uit

H0

with only two periods t

= 1, 2

: 11 = 12

compute the F test

it is possible to test change in the slopes of several variables


simultaneously

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Two-period Panel Data


panel data can be used to address more potential problems than
pooled cross-sections

it = 0 + 1 experit + 2 (experit ) + ai + uit


ai is not observed and is individual specic
for example, assume that ai captures the combined

wages

eect of

unobserved human capital and ability

it

is the usual error term


represents unobservable time-invariant individual variation

(heterogeneity)
individuals are esentially dierent, and that is partly why they
make dierent choices

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Unobservable Heterogeneity and Pooled Cross-Sections

wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a cross-section of individuals


ols wages const exper exper 2

i , exper it ) 6= 0

if cov (a

robust

then OLS will be inconsistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Unobservable Heterogeneity and Pooled Cross-Sections

wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a cross-section of individuals


ols wages const exper exper 2

i , exper it ) 6= 0

if cov (a

robust

then OLS will be inconsistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Unobservable Heterogeneity and Pooled Cross-Sections

wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a cross-section of individuals


ols wages const exper exper 2

i , exper it ) 6= 0

if cov (a

robust

then OLS will be inconsistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Unobservable Heterogeneity and Pooled Cross-Sections

wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a cross-section of individuals


ols wages const exper exper 2

i , exper it ) 6= 0

if cov (a

robust

then OLS will be inconsistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Getting Rid of Unobservable Heterogeneity


wages

it = 0 + 1 experit + 2 (experit ) + ai + uit

suppose that we have a panel with T

=2

we can look at the changes observed in each individual


this is referred to as taking rst dierences:

wagesit = wagesit wagesi ,t 1


by taking rst dierences, we get rid of a

wages it = 1 exper it + 2 exper 2it + uit


if cov (u

it , xit ) = 0

then OLS in rst dierences will be

consistent

R. Mora

Cross-sections and Panels with Two Periods

Pooling Cross-Sections Across Time


Modelling Time Eects in Cross-Sections
Chow Tests of Structural Change
Panel Data
Summary

Summary

pooling cross-sections is useful for improving accuracy and also


to test time eects
it is very easy to test structural changes in pooled cross
sections
with panel data we can do everything we do with pooled
cross-sections
in addition, we can also obtain OLS estimates which are robust
to unobservable heterogeneity

R. Mora

Cross-sections and Panels with Two Periods