 |
 |
Wednesday, Nov 25
|
|
|
|
User Guide Tutorial #5-A
Intergenerational Analysis Using the Panel Study of Income Dynamics (PSID) --
With an Application to Homeownership
Ngina Chiteji and Frank Stafford
December, 2003
I. Introduction
In recent years there has been
growing interest in exploring connections among family members, particularly
connections that span across the generations.
In both research and policy circles there has been interest in a range
of issues, from father-son correlations in occupation, educational attainment,
and earnings (Chadwick and Solon, 2002; and Solon, 1992, for example); to
mother-daughter correlations in fertility decisions and welfare receipt (An,
Haveman and Wolfe, 1993 for example); to the influence that parent's income,
wealth and socio-economic status has on their children's well-being (Conley,
1999 for example); to intra-family transfers and parent-child similarity in
asset ownership and wealth levels (Charles and Hurst 2002, and Chiteji and
Stafford, 1999, for example). Because
of its unique design, the Panel Study of Income Dynamics (PSID) is one of the
premier datasets for conducting intergenerational analyses. It contains extensive economic and
socio-demographic information about families and their relatives; and, because
it is a longitudinal survey that has followed families and their offspring
since 1968, it contains information on partial and complete life histories for
each family/individual, over different stages of their lives. In Tutorial 5 we present two exercises that
illustrate ways to access data resources when conducting intergenerational
research using the PSID. While not
exhaustive (since the PSID includes a wide variety of data elements and
therefore lends itself to a multitude of applications), the sample exercises do
serve to demonstrate how PSID data can be used to answer intergenerational
research questions in a comprehensive yet fairly easy fashion.
The two sample exercises presented
focus on the use of PSID data to examine cross-generational correlations in
homeownership, and similarities in health conditions. The first exercise, presented here in
Tutorial 5-A, shows how one can use
the PSID to put a two-generation dataset together to examine adults with their
parents in prior years of the PSID. The
second, presented as a separate tutorial (5-B), illustrates a
three-generational analysis using PSID data.
II. Overview of
the PSID
Why is the PSID an ideal dataset
for exploring connections between generations?
First, the PSID is nationally representative and contains a vast amount
of information about the social and economic characteristics of U.S.
families and the individuals within them.
The focus of the data is economic and demographic, with substantial
detail on income sources and amounts, employment, family composition changes
and housing, but since the 1980s there also has been an extensive set of health
and well-being measures which can be used to portray the full life course and
intergenerational connections. Extensive
information about asset-holdings and wealth also is available in several years
(in the 1984, 1989, 1994, 1999 and 2001), for example, and additional
information about the quality of individuals’ and families’ lives, such as
information about charitable giving and volunteer time is also available in several
years.
The second reason that the PSID
data are ideal for conducting cross-generational analyses is that the PSID
follows individuals as they reside in ever-changing families over time, along
with the new families that emerge as the children in the original families age
and move on to set up their own households.
This means that one is able to obtain a wealth of information about
different generations of a given family tree in the same dataset. In fact, because the PSID has been following
families and individuals since 1968, in some applications the files can be used
not just to put information about children and their parents together, but to
create a file of information across three generations of a family (the
children, their parents and the grandparents).
The PSID currently obtains
information on about 8,000 families and has collected information on about
65,000 individuals. It therefore
supports possible analyses of the U.S.
population as a whole, or of special sub-groups. One can decompose the population by race or
gender for example, or across different generations, and often still have
sample sizes that are large enough to draw reliable conclusions about families'
circumstances and their behavior.
III. Sample
Exercises Demonstrating the Way the PSID can be used in Intergenerational
Analyses
How might one put together a
dataset with information about parents and their children, and what type of
question might one answer with such data?
How one puts an intergenerational dataset together can depend upon the
nature of the question that one is asking.
Some common challenges emerge regardless of what type of analysis one is
doing, however.
The first issue any researcher will
face is the need to link parents with their offspring. When examining dependent children, this task
is facilitated by the PSID’s inclusion of information that uniquely identifies,
for a given year, both the family that a child lives in and the head and, when
relevant, the wife of that family within the files containing detailed
information about children. Researchers
interested in conducting extensive analyses of dependent children might choose
to use the PSID’s 1997 Child
Development Supplement (CDS), for example.
As noted in Tutorial
4, the CDS contains a wealth of information about children and the
different settings in which their lives unfold, such as their school environment
and neighborhood context; and it also provides basic family-level information
for the family in which the child resides, such as total family income and the
employment status of the head and wife of that family, who are typically the
child’s parents, for the year that the survey was done. The CDS also lists an identification number
for the family in which the child resides, along with the unique identifiers
for the primary caregiver of the child (typically the mother) and for each
child’s second caregiver (often the father), and a researcher can use these
identification variables to attach detailed information about the child’s
parents or its family, obtained from the PSID’s annual family file, to each
child in the CDS. The process of linking
offspring and parents is therefore fairly straightforward in this case, since
one is able to obtain information about children along with some information
about their caregivers (again, typically the parents) in one location, and
because the CDS gives the user the information that is needed to locate other
information about the child’s family and parents elsewhere in the PSID (in the
yearly or 'core' family and individual files for example). Tutorial
4 offers an example of the way that CDS data can be used, and how
one can match information about children with information about a parent. Tutorial 5-B uses the CDS data to conduct an
analysis that also merges in information about children's grandparents.
Constructing a dataset on parents
and their adult offspring is slightly more complicated. In this instance a researcher is required to
match parents and their kids together by tracing the adult children back to
their families of origin. This step is
facilitated by the fact that the individual files at the PSID Data Center give the identification
numbers for the families in which an individual resided in ever year that an
individual appears in the PSID along with information about the individual’s
relationship to the head of this family (in the annual individual files), and
by the fact that the Data Center also lists two key identification variables
for the mother and father of each individual that appears in the dataset (under
the "sampling variables, birth and marital variables" data group
listed at the Data Center). These two
variables are the 1968 family identification number and the 1968 person
number--key variables in the PSID because they allow one to uniquely identify
each individual in the study. One can
use the two to create a (composite) identification variable unique to each
individual, and this variable can then be used to identify the proper records
of information for each individual in the PSID from different years.
Below,
we demonstrate how one can subset the PSID data to obtain a customized dataset
with information about parents and adult children from different decades. The dataset will then be used to explore the
following research questions: Is there a
connection between parents' homeownership behavior and that of their
children's? That is to say, do
individuals who grew up in families in which the parents owned their home have
a greater tendency to go on to own their own homes than individuals who were
not similarly "exposed" to home-ownership when they were dependent
children? We then provide a discussion
of the steps one might take to create a customized dataset with information
about parents and adult children in the same year. And, in tutorial 5-B, a companion piece to
this tutorial, an exercise is presented to show how the PSID data can be used
to analyze connections across three generations of the family tree. This exercise focuses on health outcomes.
IV. Analyzing Homeownership Across the
Generations
The objective of the following
exercise is to obtain the data necessary to compute the rates of homeownership
for young adults with different family background characteristics. As shown in Table 1, we will be interested in
determining if a greater fraction of individuals who grew up in families with
parents who owned their own home go on to own homes during adulthood than those
whose parents were renters. As part of
this we will calculate the percentage of young adults who come from a
background in which their parents did not own their home but go on to own their
home nonetheless. Filling in Table 1
allows the user to determine whether those whose parents exposed them to
homeownership appear to be different from those who did not. (While economists typically are interested in
explaining economic phenomenon in addition to describing patterns, we will
leave the analysis needed to explain any gaps in homeownership rates that we
find for the interested and enterprising user to do on his or her own.)
Table
1. Comparison of homeownership rates by whether one’s parents
owned their home or rented
|
|
All young families
|
Young families of
individuals whose parents owned their home during their childhood
|
Young families of individuals whose parents did not own
their home during their childhood
|
|
Percent owning their
home
|
?
|
?
|
?
|
What data will we need?
We will
need to obtain data about the housing situation of individuals "now"
and housing of their parents in "prior years." A series of housing questions has been asked
each year in the PSID, 1968 to the present.
For example, respondents are asked whether they own their residence or
whether they rent it, how much rent they pay, whether they live in public housing,
and, for homeowners, the value of their home in terms of what it could be sold
for. To view sample housing questions
see the housing section (Section A) of the PSID questionnaire at http://psidonline.isr.umich.edu/Data/Documentation/cai_doc/2001_interview_year/Section_A____Housing.htm
or, see ftp://ftp.isr.umich.edu/pub/src/psid/questionnaires/q71.pdf for 1971.
In this tutorial we will use the responses to questions A19 in 2001 and
C2 from 1971 (whether the individual owns or rents). As you will observe from examining the
questionnaire, however, there is much other additional information about
housing for researchers to use.
How and where will we get the
data?
We will
restrict our analysis to young adult families.
We will define these families as those in which the head or the spouse
is an individual in the 30 to 38 age range as of 2001. We put an age restriction on our sample for a
few reasons. First, because we want
individuals who are old enough to be out on their own, i.e. to be living
independently or in families in which they are either heads or wives, and
individuals who also are mature enough to be in the portion of their life cycle
in which they would be "settled" enough for homeownership to be a
reasonable option for them. If we think
about the universe of people usually considered to be adults in U.S.
society, we have anyone age 18 or older.
However, when one thinks carefully about the situation of some of these
individuals, it is clear that not everyone above the age of 17 would be
expected to be in circumstances in which homeownership would be sensible. College students, for example, are not
expected to be homeowners since many of them live in dorms and since many are
only temporarily residing in the locations in which they are attending college. Similarly, it typically takes college graduates
a while to become settled in their careers and to enter into a life situation
in which they are likely to want to own a home.
We also want to keep our sample
from getting too large to be easily manipulated in Excel, since the tutorial
will be using this program for data analysis.
Importantly, we also need to ensure that our individuals are young
enough to be offspring of the original PSID sample families, so that
information about their parents will be available. This is the rationale for putting an upper
bound on the age range. (Of course, for
researchers who are experienced users of Excel, or for those who plan to use
SAS, SPSS, or Stata to analyze their data, there is not necessarily a need to
arbitrarily limit the age range and size of the dataset as we are doing in this
tutorial.)
What kinds of information do we
want about our young adults? For our
analysis we need information about their homeownership status, and information
about their parents' homeownership status during the young people's childhoods.
In addition, we will include the
family income of the parents in 1971 and family income of the adult children in
2001. This will allow a check of whether
any observed home-ownership pattern across the generations may be simply
reflecting an income correlation across the generations. A full analysis would be more involved and
allow for the fact that income, particularly normal or permanent income of
families, is a strong predictor of home ownership and housing consumption.
We will get the information about our young adult families from the 2001
files. To obtain the information about
parents in a simple and straightforward fashion, we will examine the parent's
housing situation 30 years prior to the adult children's (a time at which the
younger generation would have been ages 0 to 8 and therefore living as
dependents in their parents homes). This
means we need data for the younger generation (the adult children) from 2001
and that we will need to connect it to data about the older generation (the
parents) from 1971. To obtain this
information we will need to draw from two types of PSID "files" or
"data groups." We will use
individual-level data for 2001 and 1971 and we will want to combine this with
family-level data from these same years.
(See Tutorial
1 for a more in-depth discussion of the data groups listed at the Data
Center.) The individual data are necessary because they allow us to obtain
identification information for the families that our young individuals resided
in during the years 2001 and 1971. The
family-level information is necessary because the housing information is
actually collected at the family-level and stored as information in the PSID
"family files." This means
that once we know the individual's family of residence in each year, we can
find out whether that family of residence owned or rented. For 1971, the family that an individual lives
in will correspond to the family that he/she grew up in. In 2001, the family will be a household in
which the individual of interest is either a head or wife, and we will be able
to exclude the thirty-somethings who have not left home and set up independent
households.
How do we match parents with
their adult offspring?
In order to use the Data Center to
obtain information about the two generations simultaneously, we will draw a few
additional variables beyond our analysis variables. As discussed earlier, when the PSID collects data
about individuals, it records the individual's "status" in the family
in which he/she resides. The PSID does
this by noting each individual's relationship to the head of the family in
which they are living. Accordingly, a
spouse of the head is listed as the "wife" and the relationship to
head variable is coded to reflect this status.
(By convention, when a family is headed by two adults, the adult male is
usually treated as the head. See Tutorial
1 for additional discussion of this matter.) Similarly, children receive codes indicating
their relationship to the head. In early
years this was "child of head," while there are separate codes to
indicate a son or daughter in later years.
An extensive list of the codes for this variable can be viewed at the Data
Center when one selects the
relation to head variable. Or, one can
go to http://psidonline.isr.umich.edu/data/Documentation to view a complete PSID codebook. We want to select the relationship to head
variable for both 2001 and 1971 when we choose variables via the Data
Center. We will use these data to make sure that our
young adults are heads or wives in 2001 (meaning they have set up independent
households of their own), and to make sure that the family identification
numbers that we obtain for 1971 represent families in which the individuals
were dependent children during 1971.
V. Getting the
Data for the Homeownership Exercise
To
construct the dataset that you need to do the analysis of homeownership, go to
the Data Center at http://simba.isr.umich.edu
and click on “variable selection.” Next,
when presented with the list of data groups under “PSID Data,” you want to
check the box for PSID individual data, the box for PSID family data, and the
box for Income Plus data.
Screenshot #1 Select
Data Group(s)
This will take you to a screen with a range of years for
individual data and a range of years for family data, each spanning back to
1968 when the PSID first began.
Screenshot #2 Select data categories and years
Screenshot #3 PSID/CDS Data Cart Contents
Note
that you can view documentation for each variable by clicking on the “view”
statement below the book icon in the variable list. This is helpful if you are unsure how a
variable is coded or if you want to see the actual question text that was asked
(when applicable or explanation text for generated variables).
Finally, note that the Data
Center automatically adds a few
variables to your cart. Because the 1968
family interview number (ER30001) and the 1968 person number (ER30002) are key
identifier variables in the PSID, the Data
Center will give you these
variables automatically. After you have
selected all your variables and viewed your variable list to confirm that you
have all the variables that you need, you want to click “Get Data and/or
Codebook.” This will take you to a
screen where the Data Center
allows you to specify the form that you want your dataset to take.
What
to do if you do not know the exact names of the variables that you want: As a quick aside you should note that if you
did not know the names of the variables that you wanted, you could still
construct your variable list at the Data
Center. The Data
Center has a feature that allows
you to type in a keyword(s) in order to identify variables. For example, suppose we did not know that
ER17043 was the variable name for the variable indicating whether a family
owned or rented its home in 2001. If
this were the case we could simply select “search and browse” instead of
clicking on “variable list” at the Data
Center’s opening page. This would take you to a screen where you can
search for variables using keywords or browse the variable list by
category. If you choose “variable
search” you are taken to a screen where you can type in keywords to conduct a
variable search. If you type the words
own and rent in the search box, and then select “and” from the “search type”
box you can instruct the Data Center to search through the list of variable
names in the family files to identify ones that contain the words “own” and
“rent." If you choose the word
"own" you would have many variables since people are asked about
owning other things and because the word "own" also appears as an
adjective in several places. (Remember
to also click beside the box for PSID family while at this screen.) After you submit this search, the Data
Center would give you a list of
variables with these two words in the name.
The first one listed comes from the 2001 family file (as indicated in
the year column to the left of the variable names). This is our familiar ER17043. (You could confirm that this corresponds to
question A19, the own or rent question, by clicking on the view command to the
right of the variable name and label.
This will open a documentation box that tells you a bit about the
variable.) Upon determining that this is
the variable that you wanted, you could simply add it to your data cart by
clicking the box under the green plus sign; this instructs the Data
Center to add this variable to your
list of desired variables.
Choosing
appropriate output options
Now
that you have reached the screen where the Data
Center asks for the format you want
your data delivered in, you should select “Microsoft Excel Spreadsheet” for the
purposes of this tutorial, and indicate whether you would like the Data
Center to create a codebook to go
with your customized dataset. (The
default option for the codebook is no.)
You also will want to instruct the Data
Center to subset your data at this
stage. This means we need to type the
following command in the subsetting criteria box:
ER30069 = 3 and ((ER33603 = 10 and ER33602 =1) or ER33603
= 20 or ER33603 = 22) and (ER33604 > 29 and ER33604 < 39)
What is the point of this subsetting command? It instructs the Data
Center to restrict your dataset to
cases in which an individual was a dependent child in a family in 1971 and a
head or wife of a 2001 family. This means
you are choosing individuals who had formed their own households by 2001, but
who were dependents back in 1971. The
last command in the string of subsetting criteria limits the age of the heads
and wives that we are drawing to the 30 to 38 age bracket (since we said
earlier that this is the age range that we wanted to work with). We also add an additional restriction when
working with the 2001 relationship to head criteria: In instances in which the individual is the
head of the 2001 family we want to make sure they are a current head, which is
why we add the restriction that ER33602 be equal to 1 for the head.
Two
final matters. First, if you want the Data
Center to e-mail you to tell you
when your dataset is ready you should enter your e-mail address in the box
provided. Second, doublecheck to make
sure that you have “All individuals” selected under the Data File Options
section of the page. (This is the
default, so the circle beside “all individuals” should be filled in
automatically.) NOW hit submit and let
the Data Center
do its work. When your dataset has been
created, you will receive an e-mail indicating that it is ready, and allowing
you to download it. If you did not
provide an e-mail address you should receive a job completion notice on the
screen before you.
VI. Using Excel on Your Output
Subset
You
should have a dataset with 13 variables (13 columns,) and 1085
observations. When working with this
dataset there are a few Excel commands that will come in handy. Table 2 lists these commands and offers a
brief explanation of each one. You may
want to use the first one so that you can scroll down through all the records
of your dataset without loosing sight of the variable names. Looking through the dataset is a good way to
verify that you subsetted correctly when you were at the Data
Center. For example, as you look down the column for
ER33603 all of your cells should contain a "3." This is because we asked the Data
Center to restrict our dataset to
cases in which we have an adult who was a dependent child (or step-child) back
in 1971. You may also want to scroll
through the dataset to make sure that the variable for the 2001 age (ER33604)
only takes on values ranging from 30 to 38.
Table 2. Some useful
Excel commands
|
Command
|
Explanation and instructions for use
|
|
“Freeze pane”
|
This command allows you to retain the image of your first
row of variable names as you scroll down through the dataset. To freeze the list of variable names in
this fashion, position your cursor in the second cell of the first column,
and then click on “windows” in the command bar at the top of the Excel file,
and then select “freeze pane.”
|
|
“Fill down “
|
This allows you to quickly copy a formula (or entry) from
one cell of a column to the remaining cells of that column. To implement the command, drag the cursor
to the bottom right corner of the cell whose entry you wish to copy. (A plus sign should appear at this point.) Then, simply left click on your mouse and
drag the cursor down to the last row of the column. When you reach the end of the column and
release your grip the entire column will be filled in.
|
|
“Copy” and “Paste Special”
|
Using these two commands in succession allows you to
convert any formulas that you have entered in a cell into the actual values
that are calculated by that formula.
To do this you first highlight the cells containing the formula
entries, then select “edit” from the command menu, and “copy” from the
resulting list. Next, return to the
“edit” entry on the command menu and select “Paste Special”. Then click on the box next to “values”
under the paste options.
|
Steps to take to perform your desired calculations
1. First, before you
begin to make any alterations to your dataset you probably want to copy your
entire dataset to a second worksheet, so that you can use this one for
calculations. (This way if you mess up
you will still have your original dataset handy.) To do this click on “edit” from the command
menu, and then choose “move or copy sheet.”
In the box that appears tab down to “move to end” to highlight that
option, and then click on the “create a copy” option. This produces a new worksheet with a second
copy of your entire dataset.
2. Delete all columns
except the ones containing the following variables: ER30067 (the 1971 interview number), V1967
(the home own/rent variable for 1971), V2226 (the 1971 family income variable),
ER33601 (the 2001 interview number for the family in which our individual
resided in the year 2001), ER17043 (the home own/rent variable for 2001),
ER20394 (the family weight variable), and FAMINC01 (the 2001 family income
variable). This leaves you with 7
columns.
3. Insert a new,
blank column beside your ER17043 column.
How do you do this?
Position your cursor at the column immediately to the right of the
ER17043 column (in column F that is).
Then select "insert" from the Excel command menu, and choose
"column" from the list of options that appears. Doing this will create a new column right
next to the ER17043 column. Why are
we doing this? As you will recall,
ER17043 is the variable indicating whether the younger generation of families
that we are interested in rented or owned their home (in the year in which we
are observing them--2001). In the PSID
this variable can take on a range of values.
A code of "1" indicates that the family owned its home, while
the PSID assigns a code of "5" for renters, and the value
"8" for any family that says that it neither owns nor rents. We want to transform this data to make it
easier to work with, because it will be easier to do the calculations if we use
the number 0 to represent instances in which the family is a renter. (We will still use the number 1 to indicate
instances in which the family owns its home.)
Accordingly, we want a new column beside the original and we want to
label it "young families own/rent 0/1" by typing this phrase in the
first row of the new column. What do
you want to do with this new column?
In cell F2, enter the following formula:
=if(E2=1,E2,if(E2=5,0,"."))
This instructs Excel to enter a one in the cell if ER17043
takes on a value of 1, or to enter a zero in column F if ER17043 takes on a
value of 5, and to enter a dot (a ".") if ER17043 takes on a value of
8. Next you want to copy this
formula into every row of column F. To
do this you can use the "fill down" command discussed earlier. Then, you may want to use the
"copy" and "paste special" commands to instruct Excel to
save the actual values in the column instead of the formula. (Remember, Table 2 tells you how to use
Excel's copy/paste special feature.) Finally,
as a last step, you want to sort your entire dataset by our new variable
and to delete any observations for which the "young families own/rent
0/1" variable takes on a non-numerical value (i.e., the instances in which
we have a "." in a row of this column). To sort the dataset by this variable,
position your cursor in the top row of this column, and then select
"data" from the command menu.
Choose "sort" from the options list that appears, and you will
see that Excel asks you what variable you want to sort the dataset by. (It should have the young families variable
listed--if not, click on the triangular icon next to whatever variable name is
listed and you will see a list of the names of all the variables in your
dataset and you can then simply scroll down until you find the variable that
you want.) Why are you sorting the
dataset by this variable? Doing such
reorganizes your dataset so that all the observations with a 0 for the newly
created variable "young families own/rent 0/1" are grouped together,
followed by the observations where that variable takes on a value of 1,
followed by the records for which the variable has a "." reported for
it. You can then eliminate these latter
cases by simply highlighting the rows containing this symbol and then choosing
"edit" from the command menu, then clicking "delete" from
the options presented, and "entire row" when prompted further.
4. Now we want to
make sure that we do not have any observations in which parents neither own nor
rent their homes. As you will recall, we
are using 1971 to obtain our background information about the adults leading
the young families that we are interested in (to gauge their exposure to
homeownership as a child, that is). The
responses for the variable V1967 ("own /rent" for 1971) therefore
correspond to information about the parents of the young families in our
dataset. Again we want to make sure that
we eliminate records in which the response was "neither own nor rent"
(i.e., instances in which the variable takes on a value of 8). We therefore want to sort the dataset by this
variable, so that all observations with the value 8 are grouped at the end of
the dataset. To sort the dataset in
this fashion, position your cursor in the first row of the column
containing the V1967 variable (this should be column B), and then select
"data" from the command menu, and choose "sort" from the
list of options that appears. As was the
case earlier, a box will appear asking what variable you wish to sort by (and
V1967 should be highlighted as the default option). After the dataset has been sorted, you
want to scroll down to the end to locate any records where this variable takes
on a value of 8. You can then delete these
observations by highlighting the associated rows, and then selecting
"edit" from the command menu, and then "delete" and then
"rows" and then "entire row" as you did before in step 3.
5. Next we want to
create a variable that isolates instances in which the parents of the adults
leading our young families (the older generation) owned their home. To do this we will create a new column that
stores such information. Position your
cursor in column G and select "insert" from the Excel command menu,
then choose "column." This
will create a new, blank column at G, and you want to label it "parents
own 0/1." In cell G2, insert the
following formula: =if(b2=1,b2,0) Then copy this formula to the remaining cells
of the column. (Remember, you can use
the "fill down" command to do this; AND you can use the
"copy/paste special" feature to convert the formulas to values after
you have copied the formula to all the rows of column G.) What is the purpose of this step? With this step you are re-coding your parental
ownership variable (the 1971 homeownership variable) so that it takes on a
value of 1 for any parents who own their home and a value of zero
otherwise. Having the responses coded in
this fashion will make it easier to compute the homeownership rate for parents
with the data.
6. Now we want to
create a variable that isolates instances in which the parents of the adults
leading our young families were renters back when our young adults of interest
were growing up. To do this we create
another new column beside the "parents own" column. To do this, position your cursor in column
H. Select "insert" from the
command menu, and then choose "column" from the options list. This will create a new, blank column at
column H. You probably want to label
this column "parents rent" to serve as a reminder that this will be a
column that tells us whether the parents were renters. In cell H2 enter the following formula: =if(b2=5,1,0)
Then copy this formula to the remaining cells of the column, and use the
copy/paste special feature to transform your formulas into values when you are
done. What are you doing with this
step? You are creating a variable that
directly indicates whether the parents were renters (with a 1 for yes and a
zero for no).
7. With steps 5 and 6
we have separated out the members of the younger generation whose parents were
owners from those whose parents were renters.
We have one more step to take before we can do any calculations,
however. Because we need to use weights
in order for the PSID to be nationally representative, we now want to create a
column that combines the weights information with the parental-ownership
information, and another column that combines the weights information with the
parental-renting information. To do this,
we will create two new columns to the right of the "parents
own" column (i.e., to the right column G).
To do this we need to position the cursor at column H. Next choose "insert" from the
command menu, and select "column" from the options list. Then repeat your actions. You should now have two new, blank columns at
H and I. Label the first one
"weighted p-ownership" (column H) and label the second one (column I)
"weighted p-renting." Now do
the following two things: (a) First, enter
the following formula in cell H2:
=(g2*k2) and then copy this formula into the remaining rows of column H,
and then use the "copy/paste special" feature to convert the formulas
into values. (b) Second, enter the
following formula in cell I2:
=(j2*k2) and then copy it to the remaining cells of column J, and then
convert the formula entries into values using the "copy/paste
special" commands.
8. Pheew…now we're
finally ready to do some number crunching.
Find some blank space in the far right section of your worksheet (somewhere
AFTER your columns containing data). To
calculate the proportion of young families that own their homes enter the
following formula in an empty cell:
=sumproduct(f2:f967,k2:k967)/sum(k2:k967)
To calculate the homeownership rate among young families
whose parents also owned their own homes enter the following formula in a blank
cell:
=sumproduct(f2:f967,h2:h967)/sum(h2:h967)
To calculate the percent of young families coming from a
background of renting (as a child) who have since gone on to own their own home
during adulthood enter the following formula in a blank cell:
= sumproduct(f2:f967,i2:i967)/sum(i2:i967)
You can now fill in
Table 1!!! The first value that you
calculated gives you the rate of homeownership among young families. The last two numbers that you computed break
this population up into two different categories based on their family
background (whether they grew up in households where their parents owned their
home or rented). This allows you to
determine whether homeownership rates differ depending upon whether one's
parents owned their home or not. As you
can see, the homeownership rate is higher for young families that come from a
background of parental ownership than it is for young families whose parents
were renters. The gap is pretty large.
9. In the previous
step we saw that the homeownership rate is higher among those whose parents
owned their home than it is for those whose parents did not. While this suggests the presence of an
association between homeownership and exposure to homeownership as a child, it
is reasonable to ask whether an alternative reason for the association
exists. For example, someone might ask
whether the apparent association between parental homeownership and
homeownership among adult children simply reflects a correlation between
parental and child incomes. The argument
would be that if high income individuals are more likely to own homes than low
income individuals, and if incomes are correlated across generations, then our
finding that homeownership is more likely among individuals whose parents were
homeowners might simply be an artifact of the cross-generational income correlation. In practice, research within economics does
suggest that there is intergenerational correlation in incomes (Solon, 1992 for
example). Accordingly, one might want to
supplement the bivariate analysis in step 8 with an investigation that takes
parental and child incomes into account.
To do this we will use Excel's regression analysis tool. It can be found by clicking on
"tools" from the command menu, and then selecting "data
analysis," and then highlighting "regression" and clicking
"OK". This will prompt Excel
to display a box in which you need to enter some additional information. Excel will ask you for an "input-Y"
range and an "input-X" range.
The column of unweighted 2001 homeownership data will be your
"input-Y" range. Before
starting however, you may want to copy your dataset into a new worksheet so
that you can use it specifically for the regression analysis. (Remember, to create a new copy of a
worksheet you choose "edit" from the command menu, and then
"move or copy sheet." Then,
you highlight "move to end" and click on the box beside the phrase
"create a copy.") Now you
are almost ready to run your regression.
Before specifying your "input-X" range, however, you want to
make sure that you have the variables that will be serving as independent
variables (the regressors) in adjoining columns. If the (1971) parents own variable, and the
2001 family income variable (FAMINC01), and 1971 family income variable (V2226)
are not presently in adjacent columns, reorganize your dataset so that they
are. (You can do this easily by
inserting blank columns beside one of the variables and then copying the other
two variables into them.) Finally, be
sure to click in the box under the word "label" since your Excel file
contains a row of labels.
10. Now run your regression and view your
regression results! When looking at
the table of regression output, you should find that the coefficient for the
parental ownership variable is positive and that its t-statistic indicates that parental ownership has a statistically significant effect even when we
take into account the influence of parents' income and adult kids' income on
adult kids' homeownership. (For those
who need a quick statistics refresher, as a rule of thumb, a t-statistic larger
than two denotes statistical significance at a generally acceptable level.)
(Some notes for the super-curious: It is not possible to run weighted
regressions in existing versions of Excel, and Excel only offers the
possibility of running ordinary least squares regressions. There has been some debate in the social
sciences literature about whether it is necessary to include weights in
regressions [DuMouchel and Duncan, 1983, for example]. While it is conventional to use weights in
regressions for many analyses of PSID data, particularly analyses that use
income measures as the dependent variable, for the purposes of this tutorial the
unweighted regression suffice to show how one might account for the influences
of other phenomenon, such as parent-child correlations in income, in order to
determine whether the association between parent and child homeownership can
really be interpreted as suggesting that growing up in a family that owns its
home has positive effects on homeownership because of the exposure that it
provides. If you want to analyze
homeownership in even more detail, you may want to add additional regressors,
and to use a different estimation technique such as probit or logistic
regressions, because there are shortcomings associated with the use of using
ordinary least squares estimation when the dependent variable is a discrete
variable such as our homeownership variable, which only takes on values of zero
or one. You probably already know this
from your statistics or econometrics course.
However, if you want a refresher, see Greene, 2002 for a discussion of
the issue.)
VII. Other
Possible Types of Intergenerational Analyses--Food for Thought
Two interesting questions that
researchers often encounter in doing intergenerational analysis is (1) whether
the researcher will require repeated observations on each generation and (2) whether
the analysis will involve time measures that are asynchronous. While we do not illustrate all of these
issues with this simple tutorial, one of the advantages of using PSID data to
do intergenerational research is that each can be dealt with in a
straightforward fashion. For example,
suppose one wanted to examine the correlation in income between our young
families above and their parents in 1997, if one wanted to know whether
families that are well-to-do tend to come from extended families that are
similarly situated? This is an example
of analysis using synchronous time measures, rather than observing each
generation at a different point in time as we did in the exercise presented
above. In this hypothetical exercise,
the best way to match adult children and their parents would be through use of
the 1968 family identification number and 1968 person number that the PSID
provides for each individual in the survey.
These two variables can be used to create a unique identifier variable
for each individual in the PSID. (This
is done by multiplying the 1968 family ID number by 1000 and then adding on the
value of the person number.) To put
parents and adult children together in this situation, you would simply instruct
the Data Center to select individuals who were heads or wives in 1997, restricting
them to lie between the ages 30 and 38 as we did above, and to output their
1997 family interview number and their 1997 family income. You also would instruct the Data Center to
provide the 1968 family ID and person number of the head's (and wife's) mother,
AND the 1968 family ID and person number of the head's (and wife's)
father. This parental identification information
is available for each individual in the PSID from the "Sampling
Variables" data group, and it would allow you to connect the individuals
who are heads or wives of 1997 families to their parents, and to merge the
information of the parents' families (such as the 1997 income of the parent
families) onto the records of the young families (those headed by individuals
age 30-38). The merge is accomplished by
noting that once you obtain a dataset with information for young families that
includes the unique identifiers for the parents of these young families, you
can then obtain a second dataset with information about PSID families and use
the unique identifiers for the heads and wives of the second set of families to
see if they coincide with the unique parental IDs for the first set of
families. In instances in which they
match you will have identified the parents of the first set of families. (Actually, the companion tutorial, 5-B, shows
how one can use the 1968 family identification number and the 1968 person
number to create a unique identifier for each individual in the PSID, and how
one can merge information from one generation onto the records of another
generation using the unique ID. Its
analysis does not involve synchronous time however.)
A seasoned researcher will realize
that one possible limitation of such the above approach to analyzing
correlations in family income across the generation is that a family's can
fluctuate from year to year, so if one truly is interested in exploring the
correlation of parents' economic status and their offspring's, one might rather
have a measure of income that is indicative of each individual's
"normal" income or a longer-run measure of income. (For those who have taken macroeconomics,
this is akin to the difference between transitory income and permanent income
that Milton Friedman stressed.) In such
a case, one would want to measure family income over a range of years, rather
than taking a one-year snapshot. This
can be done with PSID data. Because the
study is longitudinal, the information about each of the families in the study
covers a wide range of years. This
allows one to put together a balanced panel that covers more than one
year. (For more information about how to
assemble a balanced panel using PSID data see Tutorial #3.)
References
An, Chong-Bum; Robert Haveman and Barbara Wolfe (1993). "Teen Out-of-Wedlock Births and Welfare
Receipt: The Role of Childhood Events
and Economic Circumstances," Review of Economics and Statistics, May,
Volume LXXV, No. 2
Chadwick, Laura and Gary Solon (2002). "Intergenerational Income Mobility among
Daughters," American Economic Review, Volume 92:1, March,
pp. 335-344.
Charles, Kerwin and Eric Hurst (2002). "The Correlation of Wealth Across
Generations," forthcoming in Journal of Political Economy.
Chiteji, Ngina and Frank Stafford (1999). "Portfolio Choices….." our AER Papers
& Proceedings Paper.
Conley, Dalton
(1999). Being Black, Living in the
Red: Race, Wealth and Social Policy in America,
University of California
Press.
Greene, William H. (2002).
Econometric Analysis.
Prentice Hall.
DuMouchel, William and Greg Duncan (1983). “Using Sample Survey Weights in Multiple
Regression Analyses of Stratified Samples,” Journal of the American
Statistical Society, 78(383):535-543.
Solon, Gary
(1992). "Intergenerational Income
Mobility in the United States,"
American Economic Review, Volume 82:3, June, pp. 393-408.
Institute for Social Research |
University of Michigan |
Privacy |
Conditions of Use