America’s Big Babies: An econometric analysis of the percentage of male adults between the ages of 18 and 34 living at home with their parents in USA Project prepared for Dr John Stinespring Md Samiul H. Dhrubo 12/11/2013
America’s Big Babies: Aneconometric analysis of thepercentage of male adults
between the ages of 18 and 34living at home with their
parents in USA
Project prepared forDr John Stinespring
Md Samiul H. Dhrubo12/11/2013
I. Introduction
The share of young adults between the age group of 24 and
34 living with parents have edged up last year despite
improvements in the economy. A new study from Pew research has
estimated that a total of 21 million young adults are living
with parents, a clear sign that effects of recession are still
lingering. “Although the media at times present a picture of
an increasing proportion of young adults living in their
parent’s home, Messineo and Wojkiewicz (2004) finds that the
increase in propensity from 1960 to 1990 for young adults age
19 to 30 to live with parents was largely due to an increasing
proportion of young adults over this time period who were
never married, or formerly married – groups that are much more
likely to reside with their parents” Kreider, M said in a
speech at the ASA annual meetings in New York, August 12,
2007. The predicted percentage of young male adults living at
home is of particular importance in determining the loss of
potential productivity faced by The United States every year.
While there is a substantial literature which examines the
1 | P a g e
home-leaving (and returning) behavior of young adults, little
work has been done to show the socio-economic reasons behind
the rising trend over the last decade.
This paper provides new empirical evidence on the
relationship between percentage of males within the age group
of 18 and 34 living at home and limited labor market outcomes,
average marriage age for young people, and rent of house to
price of house ratio in the United States. I use this evidence
to argue that percentage of males between the age group of 18
and 34 are affected by these key socio-economic variables. To
understand the relationship it is necessary to understand the
uncertainties and opportunities that exist for young adults in
the labor market. I based my research paper primarily on one
paper, written by Liu, Yang, Di Zhu “Young American Adults
living in Parental Homes,” (2011). The data I used was for the
United States as a country dating back to 1983 through 2012
and primarily sourced from The Bureau of Labor Statistics
(BLS), The Current Population Survey (CPA) and American
Housing Survey (AHS).
2 | P a g e
II. Literature review
The basis of my research and calculations are from a
paper entitled “Young American Adults living in parental
homes” written by Zhu Xiao Di, Yi Yang and Xiaodong Liu. Their
paper, written in 2002 reviewed the literature of young adults
(ages 25-34) living in parental homes in regard to gender
difference, racial difference, family structure variation,
parental resource gap, personal income gap, and the long-term
trend. They test to see the effect of personal income,
parental resource, and race on the living arrangements of
young adults. They based their research on data collected from
The Current Population Survey (CPS). One of the limitations
they faced while using CPS data for their analysis is that the
data did not have information on rent. To amend, they
generated a median monthly contract rent variable based on the
American Housing Survey (AHS) of 1999 which was adjusted for
four regions and metropolitan status, namely inner cities,
suburbs, and non-metro areas. For each dataset, they estimated
the effect of various factors on the probability of young
3 | P a g e
adults living at parental homes, controlling for selected
demographic young adults living at parental homes, controlling
for selected demographic characteristics. Their dependent
variable is whether the young adult lives in parental home
(1=yes, 0=no). Independent variables include young adults’
personal income, average rent in an area (For CPS 2001 data),
parental resources (for PSID data), age, gender, race,
educational attainment, marital status, and regional and urban
variation (for PSID data). Their analysis confirms as pointed
out in their research by Liu, Yang, Di Zhu (2002) “their
belief that personal income is one of the most important
factors explaining the living arrangements of young adults
(ages 25-34)” (p. 40). Controlling for parental resources and
selected demographic factors, those with lower personal income
are more likely to live in parent’s home. Even though their U-
shaped pattern representing the long term trends of co-
residence was in line with the overall economic conditions in
income distribution such as family income inequality, low-wage
share of total employment, inequality in wages and salaries,
and the number of persons below the poverty level, their
4 | P a g e
conclusions do not give us a numerical prediction of the
percentage of big babies living at home.
The Liu, Yang, Di Zhu paper wasn’t the only paper used in
my research, but the theory served as the back bone of my
model. The strong relationship between living arrangements and
personal income encouraged me to observe what other socio-
economic factors can affect this relationship and change the
percentage of young adults living at home. Another piece used
as a reference was published by the Fertility and Family
Statistics Branch, U.S. Census Bureau and presented by Rose M.
Krieder in August of 2012 and was titled “Young Adults Living
in Their Parent’s Home”. Her literature “Young Adults Living
in Their Parent’s Home” (2007) examined how “… [T]he
characteristics of young adults living in their parents’ home
might differ from young adults living elsewhere” (p.1).
Krieder’s findings indicate that the profile of young adults
living in their parents’ home suggests that young adults often
live in their parents’ home for their own benefit. Another
paper I found interesting was, “Intergenerational Transfers
and Household Structure Why Do Most Italian Youths Live With 5 | P a g e
Their Parents?” by Marco Manacorda and Enrico Moretti (2002).
I did not focus entirely on this due to the geographical
relevance and it used independent variables which are
primarily social factors that are typical characteristics of
Italian Youths only. Their basic analysis was that Italy is an
outlier in terms of the living arrangements of its young man.
III. Methodology and Data
To test the hypothesis that socio-economic factors have a
greater impact on percentage of young adults (18-34), I
created a total of three linear-logged model based on the
model used in the Liu, Yang, Di Zhu paper, but I added
different independent variables and expressed my dependent
variable as a percentage of young adults (18-34) living at
home. My models attempt to estimate the impact of socio-
economic factors including rent to price ratio of housing,
labor market participation ratio of people over the age of 65,
average marriage age of males and real weekly wage of adults
(18-34). Percentage of young adults living at home denoted by
PHt, rent to price ratio of housing as RPt, average marriage
6 | P a g e
age of males as AVGMt, labor market participation ratio of
people over the age of 65 as LPt, and real weekly wage as RWt
all of which I expect to have a significant impact on
percentage of young adults living at home.
The idea to start with a linear-logged model came from
Liu, Yang, Di Zhu paper and I also thought the variables
should have a linear-logged relationship with my dependent
variables and the errors to be normally distributed. I have
run a Jarque-Bera Normality test to show that my errors are
normally distributed. I decided to log some of the independent
variables in the model because of the fact that the regressand
and some of the regressors are in different units. Logging
some of the regressors will help me minimize the spread of the
data and attempt to get the data on a comparable scale.
The linear-logged model is written as follows:
Model I: PHt = β0 + β1 Log (RPt) + β2 Log (AVGMt) + β3 LPt + β4 Log (RWt) +Ut
From this point I developed my second model which is
essentially the same as my original model but controlling for the
7 | P a g e
independent variable, labor force participation ratio of people over the age
of 65(LPt).
Firstly, the introduction of this control variable will
enable me to predict the long-run trend of the percentage of
young adults (18-34) living at home without taking into
consideration a phenomenon which has been only recently observed
in the labor market and might not hold in the long-run with the
economy emerging out of the Great Recession.
Secondly, another reason behind dropping the independent
variable, labor force participation ratio of people over the age of 65(LPt) in
the second model is entirely based on suspecting
multicollinearity between LPt and one or more independent
variables such as average marriage age of males as people are
less likely to get married if they do not have a stable job. I
included tests results in the appendix section to show
evidence of multicollinearity.
The second model is as follows:
8 | P a g e
Model II: PHt = β0 + β2 Log (RPt) + β3 Log (AVGMt) + β4 Log (RWt) + Ut
Building up on this model a dummy variable was added to
make a third model and to account for any impact that a
recession may have on real wages and coincidentally affect the
dependent variable, percentage of young adults (18-34) living
at home. We added this variable on the account that recessions
would have a qualitative impact on percentage of young adults
(18-34) living at home, one that couldn’t be measured by
adding numerical data. The third model is still controlling
for the independent variable, labor force participation ratio of people
over the age of 65(LPt):
Model III: PHt = β0 + β1 Log (RPt) + β2 Log (AVGMt) + β3 Log (RWt) + β4
(Recession*Log (RWt)) +Ut
All models underwent a series to test to verify their
legitimacy and to ensure no models contained underlying
problems, resulting in biased predictions. The first test was
for normality which was done by looking at the probability of
the Jarque-Bera Normality Test. It is important that the error
terms u are normally distributed. In the classical normal
9 | P a g e
linear regression model (CNLRM) it is assumed that the error
terms follow the normal distribution (with zero mean and
constant variance). Using the central limit theorem (CLT) to
justify the normality of the error term, I was able to show
the OLS estimators themselves are normally distributed. This
in turn allowed us to use the t and F statistics in hypothesis
testing in small, or finite, samples like my samples.
Therefore the role of the normality assumption is very
critical. Due to the small size of the samples I ran a Jarque-
Bera Normality Test and it showed that the errors were
normally distributed. All the Jarque-Bera Test error terms
output gave me probabilities of more than 20% as shown in the
table below:
10 | P a g e
Figure 1
The errors are normally distributed because the
likelihood of getting a Jarque-Bera score of 1.5654
(approximately) and the errors being normally distributed is
45.72% (approximately).
Secondly, I tested the slopes of the regression line to
see if there is a significant relationship between the
independent and dependent variable. Just because the slope
coefficients are not equal to zero, it doesn’t mean that there
is a statistically significant relationship. To evaluate I
conducted t-tests for each slope coefficients of the
independent variables. All my slope coefficients had t-
statistics greater than the t-critical value at 5%
11 | P a g e
significance level. So I rejected the null hypothesis and I
could statistically conclude that there is a relationship
between the independent and dependent variables. My F-
statistic computed was also greater than F-critical which
determined that there is a significant relationship between
the dependent variable and any of the independent variables in
our model. The adjusted R2 of all three of my models were
high, which was an excellent indicator that our regression
line was much better than simply using the average value of
the dependent variable for prediction purposes.
Next was to test for multicollinearity this was conducted
in many steps. My first model showed low t-stats for some of
the independent variables with high probability and R2 and F-
statistics were high which were good signs indicating that the
model suffers from multicollinearity. After noticing that two
of my independent variables Log RPt and LPt were showing low t-
stats with high probability I conducted a simple pair-wise
correlation test and it confirmed that the independent
variables Log AVGMt is highly collinear with Log RWt and LPt as
shown in the table underneath.12 | P a g e
LOG(RWt) LOG(RPt) LOG(AVGMt) LPt
LOG(RWt) 1.000000-0.531632 0.835753 0.746151LOG(RPt) -0.531632 1.000000 -0.318637 -0.411232LOG(AVGMt) 0.835753-0.318637 1.000000 0.906715
LPt 0.746151-0.411232 0.906715 1.000000Table 1
I further confirmed my doubts by using confidence
ellipses to decipher which variables had a possibility of
being collinear, which was indicated by an elliptical shape as
opposed to a circular one, where a circular shape would have
indicated no multicollinearity. It confirmed my simple-pair
wise correlation test.
I regressed the independent variables on the other
independent variables. Upon regressing Log AVGM on Log RW I
got an auxiliary regression R2 = 0.698483 which is less than
the adjusted R2 of the original model indicating that there is
no problematic collinearity between these two independent
variables. However, upon regression Log AVGMt on LPt I got an
auxiliary regression R2 of 0.822132. Using Klein’s Rule of
Thumb I can conclude that there is high collinearity between
the two independent variables.
13 | P a g e
I had two options for correcting the multicollinearity.
According to O.J. Blanchard, Comment, Journal of Business and
Economic Statistics, multicollinearity is essentially a data
deficiency problem (micronumerosity). Faced with
micronumerosity I decided to drop the independent variable LPt
in my second model although carefully checking for
specification bias. Even though economic theory suggest that
the labor force participation of people over the age of 65 is
important, out limitation in having a priori information on
how much it will affect the dependent variable I dropped the
variable. It corrected for multicollinearity in the first
model and the ensuing models.
Since all my data was collected for the same population
over a period of time the variables were of similar orders of
magnitude, as a result of which I did not face any trouble
with Heteroscedasticity.
The next test was for autocorrelation. I started
detecting for autocorrelation by plotting the residuals
against time, the time sequence plot as show overleaf:
14 | P a g e
Figure 2
Examining the time sequence plot as above, I observed
that our estimated error terms exhibit a pattern (negative
runs to the positive runs) suggesting that perhaps our error
terms are not random.
Then we conducted a Durbin-Watson test to check if the d-
statistic shows results of autocorrelation. The calculated d-
statistic for all our models were close 2.00 indicating there
is no autocorrelation. To avoid some of the limitations of the
Durbin-Watson d test for autocorrelation, I also used a
15 | P a g e
Breusch-Godfrey (BG) Test to further verify our observation.
Using Breusch-Godfrey test we fail to reject the null
hypothesis of no auto correlation. I also checked each model
more model misspecification using a Ramsey RESET test.
As mentioned earlier our data includes percentage of
young adults (18-34) PH, Rent to Price of houses ratio (RP),
average marriage age of males (AVGM), labor market
participation ratio of people over the age of 65 (LP), real
weakly wage (RW), and a dummy variable indicating recessions.
All the data was collected primarily from The Bureau of Labor
Statistics (BLS), The Current Population Survey (CPA) and
American Housing Survey (AHS) expressed annually between the
years or 1983 to 2012. I would also like to acknowledge
Associate Professor John Stinespring, on his contribution with
reliable dataset for years 1983 to 2011. The descriptive
statistics of all our variables is listed below.
PH RP AVGM LP RW
Mean14.76
14.699
26.84
3
13.30
0572.952
Median 14.54 4.938 26.85 12.23 588.375
16 | P a g e
8 0 8
Maximum18.69
55.327
29.10
0
17.80
0689.000
Minimum12.85
93.098
25.40
0
10.77
5419.250
St. Dev. 1.400 0.628 0.923 2.279 88.305
Skewness 1.267-
1.3760.480 0.853 -0.222
Kurtosis 4.730 3.800 2.983 2.274 1.512
Jarque-
Bera
11.76
5
10.26
21.152 4.296 3.014
Probabili
ty0.003 0.006 0.562 0.117 0.222
Sum442.8
34
140.9
79
805.3
00
399.0
10
17188.5
50
Sum Sq.
Dev.
56.85
3
11.08
2
24.71
4
150.6
44
226135.
000
Observati
ons30 30 30 30 30
Table 2
17 | P a g e
IV. Results
The results of my models showed different beta values for
each variable depending on the model.
The first model had betas were all statistically significant and
had a good Durbin-Watson score of almost 2.00.
PHt = -179.3907 + 1.658688 Log RPt + 77.72824 Log AVGMt – 0.206411 LPt -9.676319 Log RWt
(-8.342) (2.079) (9.279) (-2.070) (-7.960)
Adjusted R2 = 0.88 DW = 2.03
After the variables in the linear model were tested for
their significance the model indicates that the average marriage
age of males and the real wage of adults are the most significant
variables with probability of being equal to zero is 0.0000. Even
though the model wasn’t represented in any of the papers I used
as reference, it confirms Rose M. Krieder’s claim that average
marriage age of males have a significant impact on their decision
to live at home with their parents. My findings also confirms
18 | P a g e
Liu, Yang, Di Zhu’s finding that personal income is one of the
most important factors explaining the living arrangements of
young adults.
The intercept fail to infer any economically significant
prediction because the percentage of young adults living at home
cannot be negative. However analyzing the other slope
coefficients, we can start interpreting them. When the rent to
house price ratio goes up by 1 percent, on average, the
percentage of young adults living at home goes up by 1.659% which
meets my apriori expectation that as rent of houses increase the
percentage of young adults living at home should increase as
well. The average marriage age of males is a highly economical
and statistically significant variable as its slope co-efficient
shows. It indicates that as the average marriage age of males
increase by 1%, on average, the percentage of young adults living
at home will increase by 77.3% hence the most important
determinant of a young adult’s decision to continue to live at
home with their parents. As I mentioned it verifies Rose M.
Krieder’s claim that average marriage age of males have a
significant impact on their decision to live at home with their
19 | P a g e
parents. As labor force participation rate of people over the age
of 65 goes up by 1%, on average, the percentage of young adults
living at home decreases by .206%. It is interesting to see that
this independent variable shows such results as it contradicts my
apriori expectation of how this variable would affect the
percentage of young adults living at home. Real wages also show
an interesting relationship indicating that as real wages go up
by 1%, on average, percentage of young adults living at home will
decrease by 9.676%.
Model II: PHt = -140.8271 + 2.295887 Log (RPt) + 63.65141 Log (AVGMt) -9.042724 β3 Log
(RWt) + Ut
(-12.356) (2.940) (12.263) (-7.242)
Adjusted R2 = 0.870 DW = 1.591
After testing for the significance of the slope coefficients we
get very similar results as model I predictions. It also confirms
Rose M. Krieder’s claim that average marriage age of males have a
significant impact on their decision to live at home with their
parents. My findings also confirms Liu, Yang, Di Zhu’s finding
20 | P a g e
that personal income is one of the most important factors
explaining the living arrangements of young adults.
The intercept once again fail to infer any economically
significant prediction because the percentage of young adults
living at home cannot be negative. However analyzing the other
slope coefficients, we can start interpreting them. When the rent
to house price ratio goes up by 1 percent, on average, the
percentage of young adults living at home goes up by 2.296% which
meets my apriori expectation that as rent of houses increase the
percentage of young adults living at home should increase as
well. The average marriage age of males is a highly economical
and statistically significant variable as its slope co-efficient
shows. It indicates that as the average marriage age of males
increase by 1%, on average, the percentage of young adults living
at home will increase by 63.651% hence once again verifying the
most important determinant of a young adult’s decision to
continue to live at home with their parents. As real wages goes
up by 1% the percentage of young adults living at home decreases
by 9.043%. 21 | P a g e
Model III: PHt = -141.1819 + 2.371887 Log (RPt) + 63.35958 Log (AVGMt) – 8.848407 Log
(RWt) -0.243939 +Ut
(-12.323) (3.000) (12.136) (-6.975)(-0.876) Adjusted R2 = 0.869 DW = 1.754
After introducing the dummy variable in our third model the
slope coefficient fails to pass the significance test suggesting
that the percentage of young males living at home are not
affected by the decrease in real wage due to recession.
After testing for the significance of the remaining slope
coefficients we get very similar results as model II predictions.
It also confirms Rose M. Krieder’s claim that average marriage
age of males have a significant impact on their decision to live
at home with their parents. My findings also confirms Liu, Yang,
Di Zhu’s finding that personal income is one of the most
important factors explaining the living arrangements of young
adults.
The intercept once again fail to infer any economically
significant prediction because the percentage of young adults
living at home cannot be negative. However analyzing the other
22 | P a g e
slope coefficients, we can start interpreting them. When the rent
to house price ratio goes up by 1 percent, on average, the
percentage of young adults living at home goes up by 2.366% which
meets my apriori expectation that as rent of houses increase the
percentage of young adults living at home should increase as
well. The average marriage age of males is a highly economical
and statistically significant variable as its slope co-efficient
shows. It indicates that as the average marriage age of males
increase by 1%, on average, the percentage of young adults living
at home will increase by 63.384% hence once again verifying the
most important determinant of a young adult’s decision to
continue to live at home with their parents. As real wages goes
up by 1% the percentage of young adults living at home decreases
by 8.864%.
V. Conclusion
After comparing two ensuing models with my original model
(Model I) we find very similar results as predicted by Rose M.
Krieder and Liu, Yang, Di Zhu papers. My slope coefficients have
23 | P a g e
only slightly changed across the model however none of the models
are hugely different from each other when it came to explaining
the variability of the predicted dependent variable. However it
is important to note that our independent variable LPt from our
first model showed a negative relationship with our dependent
variable suggesting that as labor force participation of people
over 65 is negatively related with percentage of people living at
home. As mentioned before this is a very new phenomenon observed
in the US economy especially since “The Great Recession”. It will
be interesting to see if this relationship changes as more and
more young adults will start looking for jobs in the next few
years with 3rd quarter economic report suggesting that two more
million jobs will be created in the next coming years. (As I
caught up with next research I found out that the jobs added on
the month of august 2014, the job number is only 142,000 compared
to jobs added in the economy in the preceding month to be well
over 250,000. This is an alarming figure and needs to be further
worked on).
Looking at the statistical significance of the model the
adjusted R2 suggest that they are all great prediction models for24 | P a g e
the percentage of young adults (18-34) living at home with their
parents. Logically it would be advisable to use the simpler model
with the fewest independent variables based on the idea of
Parsimony, however I would suggest using Model I as it captures
more independent variable giving a higher Adjusted R2. Also the
slope coefficient of LPt is statistically significant. I would also ignore
the effect of recession in predicting the percentage of young
adults living at home as the slope coefficient is not
statistically different from zero at the 5% significance level.
If I had more time to allocate to this study we would be
able to identify more socio-economic indicators affecting the
young adults decision to stay at home and sacrificing
independence of living alone. At this point more research needs
to be conducted before a conclusion can be made as to whether
what other socio-economic factors have an influence on the
percentage of young adults (18-34) living at home.
25 | P a g e
Appendix
Section I:
Slope significance test for Model I:
Dependent Variable: PHSample: 1983 2012Included observations: 30
VariableCoefficien
t Std. Errort-Statistic Prob.
C -179.3907 21.50475 -8.341911 0.0000LOG(RP) 1.658688 0.797686 2.079375 0.0480LOG(AVGM) 77.72824 8.376944 9.278829 0.0000
LP -0.206411 0.099724 -2.069826 0.0490LOG(RW) -9.676319 1.215673 -7.959638 0.0000
R-squared 0.900528 Mean dependent var 14.76113Adjusted R-squared 0.884613 S.D. dependent var 1.400155
S.E. of regression 0.475614 Akaike info criterion 1.502592
Sum squared resid 5.655217 Schwarz criterion 1.736124
Log likelihood -17.53887 Hannan-Quinn criter. 1.577301
F-statistic 56.58197 Durbin-Watson stat 2.028597Prob(F-statistic) 0.000000
H0 : β1 = β2 = β3 = β4 = 0
H0 : β1 = β2 = β3 = β4 ≠ 0
α = 5%
Degree of freedom (d.f.) = n-k = 30-4 = 26
tcrit5% = 2.056
26 | P a g e
All our |tstat| >|tcrit|, so we reject the null hypothesis and we
can be at least 95% confidence that our estimated betas are
statistically significant from zero.
Slope significance test for Model II:
Dependent Variable: PHSample: 1983 2012Included observations: 30
VariableCoefficien
t Std. Errort-Statistic Prob.
C -140.8271 11.39716 -12.35633 0.0000LOG(RP) 2.295887 0.780982 2.939743 0.0068LOG(AVGM) 63.65141 5.190515 12.26302 0.0000LOG(RW) -9.042724 1.248598 -7.242301 0.0000
R-squared 0.883482 Mean dependent var 14.76113Adjusted R-squared 0.870038 S.D. dependent var 1.400155
S.E. of regression 0.504760 Akaike info criterion 1.594097
Sum squared resid 6.624336 Schwarz criterion 1.780923
Log likelihood -19.91145 Hannan-Quinn criter. 1.653864
F-statistic 65.71393 Durbin-Watson stat 1.591340Prob(F-statistic) 0.000000
H0 : β1 = β2 = β3 = 0
H0 : β1 = β2 = β3 ≠ 0
α = 5%
d.f. = n-k = 30-3 = 27
27 | P a g e
tcrit5% = 2.052
All our |tstat| >|tcrit|, so we reject the null hypothesis and we
can be at least 95% confidence that our estimated betas are
statistically significant from zero.
Slope significance test for Model III:
Dependent Variable: PHSample: 1983 2012Included observations: 30
VariableCoefficien
t Std. Errort-Statistic Prob.
C -141.1575 11.45478 -12.32303 0.0000LOG(RP) 2.366142 0.788595 3.000455 0.0060LOG(AVGM) 63.38360 5.222882 12.13575 0.0000LOG(RW) -8.863524 1.270809 -6.974712 0.0000RECESSION -0.243939 0.278525 -0.875826 0.3895
R-squared 0.886951 Mean dependent var 14.76113Adjusted R-squared 0.868863 S.D. dependent var 1.400155
S.E. of regression 0.507036 Akaike info criterion 1.630542
Sum squared resid 6.427134 Schwarz criterion 1.864075
Log likelihood -19.45812 Hannan-Quinn criter. 1.705251
F-statistic 49.03568 Durbin-Watson stat 1.754079Prob(F-statistic) 0.000000
H0 : β1 = β2 = β3 = β4 = 0
H0 : β1 = β2 = β3 = β4 ≠ 0
α = 5%
d.f. = n-k = 30-4 = 26
28 | P a g e
tcrit5% = 2.056
Since all our |tstat| are not greater than |tcrit| so we fail to
reject the null hypothesis and we can be at least 95% confidence
that our estimated betas are statistically significant from zero.
Section II:
Pair wise test for model, confidence ellipse and auxiliary
regression results:
LOG(RP) LOG(AVGM) LP LOG(RW)LOG(RP) 1.000000 -0.318637 -0.411232 -0.531632LOG(AVGM) -0.318637 1.000000 0.906715 0.835753
LP -0.411232 0.906715 1.000000 0.746151LOG(RW) -0.531632 0.835753 0.746151 1.000000
29 | P a g e
Auxiliary regression of Log (AVGMt) on LPt
Dependent Variable: LOG(AVGM)Sample: 1983 2012Included observations: 30
VariableCoefficien
t Std. Errort-Statistic Prob.
C 3.108717 0.016111 192.9586 0.0000LP 0.013589 0.001194 11.37630 0.0000
30 | P a g e
R-squared 0.822132Adjusted R-squared 0.815780
Auxiliary regression of Log (AVGMt) on LPt
Dependent Variable: LOG(AVGM)Sample: 1983 2012Included observations: 30
VariableCoefficien
t Std. Errort-Statistic Prob.
C 2.150726 0.141433 15.20670 0.0000LOG(RW) 0.179642 0.022305 8.053795 0.0000
R-squared 0.698483Adjusted R-squared 0.687714
Section III:
31 | P a g e
Model I:
Breusch-Godfrey Serial Correlation LM Test:
F-statistic 0.441716 Prob. F(3,22) 0.7255
Obs*R-squared 1.704359 Prob. Chi-Square(3) 0.6360
Test Equation:Dependent Variable: RESIDSample: 1983 2012Included observations: 30Presample missing value lagged residuals set to zero.
VariableCoefficien
t Std. Errort-Statistic Prob.
C -4.502255 22.99093 -0.195827 0.8465LOG(RP) -0.265268 0.860940 -0.308115 0.7609LOG(AVGM) 2.671937 9.164657 0.291548 0.7734
LP -0.021390 0.105820 -0.202139 0.8417LOG(RW) -0.568701 1.374306 -0.413810 0.6830RESID(-1) -0.086486 0.221765 -0.389987 0.7003RESID(-2) -0.197194 0.224157 -0.879716 0.3885RESID(-3) -0.190655 0.230274 -0.827946 0.4166
R-squared 0.056812 Mean dependent var -1.66E-14Adjusted R-squared -0.243293 S.D. dependent var 0.441597
S.E. of regression 0.492394 Akaike info criterion 1.644102
Sum squared resid 5.333933 Schwarz criterion 2.017755
Log likelihood -16.66153 Hannan-Quinn criter. 1.763637
F-statistic 0.189307 Durbin-Watson stat 2.063198Prob(F-statistic) 0.984705
32 | P a g e
Model II:
Breusch-Godfrey Serial Correlation LM Test:
F-statistic 0.501605 Prob. F(3,23) 0.6849
Obs*R-squared 1.842268 Prob. Chi-Square(3) 0.6058
Test Equation:Dependent Variable: RESIDSample: 1983 2012Included observations: 30Presample missing value lagged residuals set to zero.
VariableCoefficien
t Std. Errort-Statistic Prob.
C 1.719738 11.93287 0.144118 0.8867LOG(RP) -0.007951 0.814401 -0.009763 0.9923LOG(AVGM) -0.680223 5.388396 -0.126238 0.9006LOG(RW) 0.083450 1.294370 0.064472 0.9492RESID(-1) 0.178384 0.208116 0.857135 0.4002RESID(-2) -0.068794 0.217944 -0.315650 0.7551RESID(-3) -0.162499 0.217271 -0.747910 0.4621
R-squared 0.061409 Mean dependent var -1.21E-14Adjusted R-squared -0.183441 S.D. dependent var 0.477939
S.E. of regression 0.519931 Akaike info criterion 1.730721
Sum squared resid 6.217543 Schwarz criterion 2.057667
Log likelihood -18.96082 Hannan-Quinn criter. 1.835314
F-statistic 0.250802 Durbin-Watson stat 1.959473Prob(F-statistic) 0.953981
33 | P a g e
Model III:
Breusch-Godfrey Serial Correlation LM Test:
F-statistic 0.154976 Prob. F(3,22) 0.9254
Obs*R-squared 0.620871 Prob. Chi-Square(3) 0.8916
Test Equation:Dependent Variable: RESIDSample: 1983 2012Included observations: 30Presample missing value lagged residuals set to zero.
VariableCoefficien
t Std. Errort-Statistic Prob.
C 0.382405 12.21984 0.031294 0.9753LOG(RP) 0.008127 0.840510 0.009669 0.9924LOG(AVGM) -0.130788 5.535181 -0.023629 0.9814LOG(RW) 0.005049 1.356372 0.003722 0.9971RECESSION 0.029965 0.320852 0.093393 0.9264RESID(-1) 0.098099 0.216814 0.452459 0.6554RESID(-2) 0.004921 0.236883 0.020775 0.9836RESID(-3) -0.117395 0.224004 -0.524075 0.6055
R-squared 0.020696 Mean dependent var -1.01E-14Adjusted R-squared -0.290901 S.D. dependent var 0.470771
S.E. of regression 0.534880 Akaike info criterion 1.809629
Sum squared resid 6.294119 Schwarz criterion 2.183281
Log likelihood -19.14443 Hannan-Quinn criter. 1.929163
34 | P a g e
F-statistic 0.066418 Durbin-Watson stat 1.947174Prob(F-statistic) 0.999406
Section IV:
Model I:
Ramsey RESET TestEquation: UNTITLEDSpecification: PH C LOG(RP) LOG(AVGM) LP LOG(RW)Omitted Variables: Powers of fitted values from 2 to 4
Value dfProbabilit
yF-statistic 1.217814 (3, 22) 0.3267Likelihood ratio 4.609060 3 0.2028
F-test summary:Sum ofSq. df
MeanSquares
Test SSR 0.805389 3 0.268463Restricted SSR 5.655217 25 0.226209Unrestricted SSR 4.849828 22 0.220447Unrestricted SSR 4.849828 22 0.220447
LR test summary:Value df
35 | P a g e
Restricted LogL -17.53887 25Unrestricted LogL -15.23434 22
Unrestricted Test Equation:Dependent Variable: PHSample: 1983 2012Included observations: 30
VariableCoefficie
nt Std. Errort-
Statistic Prob.
C -36549.10 83177.65 -0.439410 0.6647LOG(RP) 330.2806 752.6735 0.438810 0.6651
LOG(AVGM) 15497.79 35286.09 0.439204 0.6648LP -41.21711 93.76256 -0.439590 0.6645
LOG(RW) -1929.148 4392.295 -0.439212 0.6648FITTED^2 -18.72183 44.26227 -0.422975 0.6764FITTED^3 0.778442 1.909809 0.407602 0.6875FITTED^4 -0.012026 0.030760 -0.390962 0.6996
R-squared 0.914695 Mean dependent var 14.76113
Adjusted R-squared 0.887552 S.D. dependent var 1.400155
S.E. of regression 0.469518 Akaike info criterion 1.548956
Sum squared resid 4.849828 Schwarz criterion 1.922609
Log likelihood -15.23434 Hannan-Quinn criter. 1.668491
F-statistic 33.69957 Durbin-Watson stat 1.997110
Prob(F-statistic) 0.000000
Model II:
Ramsey RESET TestEquation: UNTITLEDSpecification: PH C LOG(RP) LOG(AVGM) LOG(RW)Omitted Variables: Powers of fitted values from 2 to 4
Value dfProbabilit
yF-statistic 1.872391 (3, 23) 0.1624Likelihood ratio 6.555385 3 0.0875
36 | P a g e
F-test summary:Sum ofSq. df
MeanSquares
Test SSR 1.300270 3 0.433423Restricted SSR 6.624336 26 0.254782Unrestricted SSR 5.324066 23 0.231481Unrestricted SSR 5.324066 23 0.231481
LR test summary:Value df
Restricted LogL -19.91145 26Unrestricted LogL -16.63376 23
Unrestricted Test Equation:Dependent Variable: PHSample: 1983 2012Included observations: 30
VariableCoefficie
nt Std. Errort-
Statistic Prob.
C 77573.20 56626.82 1.369902 0.1839LOG(RP) -1231.602 898.7099 -1.370411 0.1838
LOG(AVGM) -34156.05 24912.76 -1.371026 0.1836LOG(RW) 4852.880 3539.113 1.371214 0.1835FITTED^2 53.73434 37.97062 1.415156 0.1704FITTED^3 -2.374822 1.630973 -1.456077 0.1589FITTED^4 0.039146 0.026168 1.495956 0.1483
R-squared 0.906353 Mean dependent var 14.76113
Adjusted R-squared 0.881923 S.D. dependent var 1.400155
S.E. of regression 0.481125 Akaike info criterion 1.575584
Sum squared resid 5.324066 Schwarz criterion 1.902530
Log likelihood -16.63376 Hannan-Quinn criter. 1.680177
F-statistic 37.10056 Durbin-Watson stat 1.814147
Prob(F-statistic) 0.000000
37 | P a g e
Model III:
Ramsey RESET TestEquation: UNTITLEDSpecification: PH C LOG(RP) LOG(AVGM) LOG(RW) RECESSIONOmitted Variables: Powers of fitted values from 2 to 4
Value dfProbabilit
yF-statistic 1.345720 (3, 22) 0.2853Likelihood ratio 5.054468 3 0.1679
F-test summary:Sum ofSq. df
MeanSquares
Test SSR 0.996551 3 0.332184Restricted SSR 6.427134 25 0.257085Unrestricted SSR 5.430582 22 0.246845Unrestricted SSR 5.430582 22 0.246845
LR test summary:Value df
Restricted LogL -19.45812 25Unrestricted LogL -16.93089 22
Unrestricted Test Equation:Dependent Variable: PHSample: 1983 2012Included observations: 30
VariableCoefficie
nt Std. Errort-
Statistic Prob.
C 56587.14 54350.69 1.041148 0.3091LOG(RP) -923.9276 887.0017 -1.041630 0.3089
LOG(AVGM) -24759.77 23758.50 -1.042144 0.3087LOG(RW) 3462.787 3322.301 1.042286 0.3086
RECESSION 95.36145 91.38539 1.043509 0.3080FITTED^2 39.46957 36.45807 1.082602 0.2907FITTED^3 -1.757769 1.569674 -1.119831 0.2749FITTED^4 0.029177 0.025237 1.156148 0.2600
R-squared 0.904480 Mean dependent var 14.76113
Adjusted R-squared 0.874087 S.D. dependent var 1.400155
S.E. of regression 0.496835 Akaike info criterion 1.662059
Sum squared resid 5.430582 Schwarz criterion 2.035712
Log likelihood -16.93089 Hannan-Quinn criter. 1.781594
38 | P a g e
F-statistic 29.75959 Durbin-Watson stat 1.876218
Prob(F-statistic) 0.000000
Bibliography
A) Liu, Yang, Di Zhu, “Young American Adults living in
Parental Homes,” (2011), Harvard University
B) Kreider, Rose, “Young Adults Living in Their Parent’s
Home”, U.S. Census Bureau, Presented at the ASA annual
meetings in NY, August 12, 2007
C) Gujarati, Damodar N., and Dawn C. Porter. Basic
Econometrics. Boston: McGraw-Hill Irwin, 2009. Print
39 | P a g e