Sunteți pe pagina 1din 9

Household Structure and Migration Conner Smith

Introduction

There are many reasons people may choose to move. Perhaps they are pursuing further

education, a better job, or a nicer neighborhood to raise kids. There is a host of other work that

examines the effects of various traits on movement such as educational characteristics of the

migrants and their destination (Waldorf, 2009), unemployment in their previous home or

prospects of higher wages in their new destination (The World Bank, 2007). This paper attempts

to unearth the effect of less tangible and obvious contributors to migration, elements of

household structure.

Anecdotally, people are generally aware of the effects that changes in household structure

have on their decisions to move. Perhaps a married couple files for divorce, and one or both

partners have to move. A parent may move in with their child in old age, or a child may move

back in with their parents due to economic hardship. People are generally informally familiar

with how these household structure variables affect their decision to move. This paper seeks to

formalize this understanding.

Specifically, this paper aims to extract a unique effect of household structure on

migration behavior to and within a city. I look at the effects of various household characteristics

(number of single father households, single mother households, households with children,

households with elderly, and average family size) on three different types of movers controlling

for various socioeconomic and demographic characteristics. I find that certain household

characteristics are indeed statistically as well as practically significant on the migratory behavior

to and within metro-areas, even under all controls. Perhaps more interestingly, I find that

household characteristics have very different impacts on different types of movers.

1
Household Structure and Migration Conner Smith

Data Analysis

Using data retrieved from the 2009-2013 American Community Survey I built a dataset

where each observation is a single metro-area (essentially cities, or groups of adjacent small

cities) with many attributes of that metro-area included as additional variables.

Figure 1: Summary statistics for migration and population variables. abroad is the number of movers from outside the U.S.,
between is the number of movers between metro areas, and within is the number of movers within the metro-area.

Figure 1 above displays summary statistics of my output variables, as well as population.

Population is controlled for in all models. Increasing the number of any kind of household also

increases the total population, which has an impact on the number of movers to and within a city.

Thus, population is an important control variable to remove large elements of bias. Figure 2

below displays summary statistics for household characteristics variables (the variables of

interest).

Figure 2: Summary statistics for variables of interest. singlefather is the number of single father households, single mother is the
number of single mother households, housewchild is the number of households with a child (less than 18 years old), housewold is
the number of households with an elderly person (over 65 years old), and avgfamsize is the average family size in that
metro-area (parent(s) and their children).

I consider socioeconomic and demographic control variables as groups. The

socioeconomic variable group includes the metro-area’s per-capita income, the median house

2
Household Structure and Migration Conner Smith

value, the unemployment rate, and the number of people who have completed a college degree

(associates or higher). The demographic controls measure racial distribution, as well as median

age. The socioeconomic controls are included to address the concern that single parent

households may occur more frequently in poorer neighborhoods, and people are less likely to

move to a poor neighborhood, thus not including them would cause bias. Demographic variables

are included to remove bias attributed to the effects of race and age both on household structure

and migration behavior.

Breusch-Pagan tests showed evidence of heteroscedasticity at any reasonable level (F-stat

≥ 17.61) for all models, and thus the standard errors of all models are heteroscedasticity-robust

standard errors. All regressions exhibit very large R​2​ values. This is more so due to the lack of

variation in the numbers of any type of movers relative to population (i.e. the number of movers

is essentially a fixed fraction of the population). Population itself is an excellent predictor of the

number of movers, and as it is included in all regressions, the regressions exhibit phenomenally

high R​2​ values. Additionally, all models suffer from very high levels of multicollinearity between

variables. As population increases, the majority of the other variables in the model (all of the

counts, such as number of singlefathers, number of people with college degrees, number of

people of each race) increase as well. This high degree of multi-collinearity will lead to inflated

standard errors, thus some variables that do not seem to be statistically significant according to

my estimates may actually have an effect in the real world.

3
Household Structure and Migration Conner Smith

(1) (2) (3)


VARIABLES abroad abroad abroad

singlefather -0.739*** -0.192 -0.237


(0.215) (0.286) (0.324)
singlemother -0.153 0.00445 0.0365
(0.136) (0.145) (0.157)
housewchild -0.0590 -0.0626 -0.0472
(0.0814) (0.0785) (0.104)
housewold -0.0342 -0.0309 -0.0229
(0.0284) (0.0253) (0.0314)
avgfamsize 1,946* 1,940 1,787
(1,012) (1,262) (1,265)

Socio-Economic NO YES YES


Demographic NO NO YES

Observations 381 381 381


R-squared 0.944 0.951 0.952
Population controlled for in all specifications.
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

Above is a table indicating all of the estimates of the effects of household characteristics

with and without the controls on the number of movers from abroad. Once socioeconomic

control variables are included, no element of household structure is individually significant. This

makes sense, as movers from abroad will be unable to observe household factors directly, and

instead will likely make decisions based on the socioeconomic characteristics of a metro-area

which are more easily observed.

4
Household Structure and Migration Conner Smith

(1) (2) (3)


VARIABLES between between between

singlefather 2.820** 3.512*** 2.173**


(1.408) (1.066) (0.991)
singlemother -0.523 -0.0525 -0.803
(0.476) (0.499) (0.681)
housewchild 0.0201 -0.152 -0.606**
(0.460) (0.397) (0.238)
housewold -0.138 -0.211 -0.479***
(0.228) (0.198) (0.135)
avgfamsize 5,902* 15,808*** 6,109**
(3,273) (4,390) (2,958)

Socio-Economic NO YES YES


Demographic NO NO YES

Observations 381 381 381


R-squared 0.859 0.884 0.934
Population controlled for in all specifications.
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

Above is a table displaying the estimated effects of household characteristics on the

number of movers from a different metro area within the United States. Unlike movers from

abroad, movers from other metro-areas do seem to be responsive to elements of household

structure within a city. The positive impact of single father is perhaps the most easily explained.

A disproportionate number of the new people arriving in a metro area from another metro area

may be single fathers, as once becoming a single father they likely have to move. The size of

these estimates also make sense, as the single father likely moves himself and his children.

Interestingly, single mother seems to have no such effect. The effect of average family size is

perhaps the most striking element. In the data, average family size only takes on values from

5
Household Structure and Migration Conner Smith

2.38 to 4.15 (see figure 2), so a one unit increase in average family size is a massive percent

increase, which likely contributes to the magnitude of the estimate on average family size.

(1) (2) (3)


VARIABLES within within within

singlefather 12.44*** 12.75*** 10.20***


(2.302) (2.170) (1.503)
singlemother 0.338 0.492 0.504
(0.781) (0.820) (0.707)
housewchild -0.133 -0.252 -1.382***
(0.540) (0.520) (0.376)
housewold -0.568** -0.620** -1.308***
(0.273) (0.262) (0.185)
avgfamsize -3,974 4,444 -4,718
(5,507) (7,707) (4,332)

Socioeconomic NO YES YES


Demographic NO NO YES

Observations 381 381 381


R-squared 0.986 0.987 0.993
Population controlled for in all specifications.
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

Above is a table displaying the estimated effects of household characteristics on the

number of movers within the same metro-area. The negative values on the number of households

with an elderly person in all specifications and the number of households with a child in the final

specification make sense, as the elderly and people with children are generally less likely to

move. While again single mother has no significant effect, the magnitude of the effect of single

father in this case is startling. A magnitude of two to three would make sense in the same context

of the effect of single father on movers from different metro-areas (the single father moves with

6
Household Structure and Migration Conner Smith

his children), however, a magnitude of about ten despite all controls seems practically significant

to city planners, and begs an explanation.

Conclusion

The results of this paper leave the possibility for much work to be done. Many variables

had unsurprising effects, such as the negative effects of households with children or the elderly

on movement within a metro-area and from other metro areas. The lack of effect of any

household characteristic variable on the movers from abroad once socioeconomic factors are

controlled for also seemed reasonable. However, the effect of average family size on movers

from different metro-areas, and the magnitude of the effect of single fathers on movers within the

same metro area remain unexplained.

At first it is simplest to attribute the effect of average family size to some cultural

element. However, it was robust to the inclusion of demographic controls suggesting it does have

a non-culturally related effect on movement. Perhaps large families tend to cluster together due

to similarity, causing large families to move to places where large families already are thus

increasing movement to metro-areas with larger families. This could occur as metro-areas with

larger families make special efforts to make the area more family friendly, which then attracts

additional large families. This suggests future work may wish to control for family amenity

related expenditure, thus possibly removing positive bias in the estimate of the effect of average

family size. This clustering effect could also take place through the medium of religion. Certain

religions may be more likely to have larger families, and people may migrate to places in which

they align with the most common religion. Either way, further investigation is needed to find

7
Household Structure and Migration Conner Smith

elements that may be biasing the estimate on average family size, and to find the channel through

which average family size has an effect.

The large effect of the number of single father households on the number of movers

within a metro-area is perhaps the most striking and almost certainly the most important finding

of this paper. The size and significance of its effect in all specifications of the model predicting

movement within a metro-area indicate that it is picking up some element of social unrest not

captured by socioeconomic or demographic characteristics. Interestingly, this element is unique

to single fathers, as single mothers have no significant effect. Given the magnitude of its effect,

the number of single fathers seems like it is most likely in part a signal of some element of social

unrest that causes movement, rather than the force behind the entirety of the movement itself.

Finding some measure of general social unrest and including it in the model may aid in reducing

the estimated magnitude of its effect and creating a more unbiased estimate of the effect of single

father, revealing that the true effect is much smaller. The friction caused by this movement

certainly has some costs associated with it, and thus warrants further inquiry.

This paper has shown that household structure does have a significant effect on movers

between and within different metro-areas in the United States, though both types of movers

respond to different characteristics of household structure. These models are far from

exhaustively including the elements that people take into account when deciding to move, so it is

very possible that the estimates of the effects of household structure are in some cases picking up

the effects of other variables, and are thus biased. However, the size of the estimates indicates

that unearthing the forces behind their effects could provide significant insight into the elements

that motivate people to migrate.

8
Household Structure and Migration Conner Smith

References
The World Bank. (2007). Migration and Remittances. Retrieved December 2015, from
http://siteresources.worldbank.org/INTECA/Resources/257896-1167856389505/Migratio
n_Chapter3.pdf
Waldorf, B. (2009). Is human capital accumulation a self-propelling process? Comparing
educational attainments of movers and stayers. ​The Annals of Regional Science,​ 323-344.

Data retrieved from…


The U.S. Census Bureau American Community Survey 2009-2013
http://factfinder.census.gov/faces/nav/jsf/pages/searchresults.xhtml?refresh=t

S-ar putea să vă placă și