This item is only available as the following downloads:
A NEW MOBILITY: INVESTIGATING THE IMPACT OF ICT ON US DOMESTIC MIGRATION PATTERNS BY REESE C CRISPEN A Thesis Submitted to the Division of Social Sciences New College of Florida In partial fulfillment of the requirements for the degree Bach elor of Arts Under the sponsorship of Dr. Richard Coe Sarasota, Florida May, 2013
i Acknowledgements I would to thank the academic guidance of Professor Richard Coe, Professor Duff Cooper, and Dr. Thomas Cooke Additionally, I would like to thank m y best friends, the friends that have shared birthdays with me, or shared secrets, or otherwise contribut ed to this thesis by nurturing my emotional health these past four strange years.
ii Table of Contents Acknowledgements .. .i Table of Abstract ... i v .. 6 3
iii List of Figures List of Tables .41 .. 42 45
iv A NEW MOBILITY: INVESTIGATING THE IMPACT OF ICT ON US DOMESTIC MIGRATION PATTERNS Reese Crispen New College of Florida, 2013 ABSTRACT Contrary to public assumption, aggregate domestic migration rates in the United States have declined steadily over the past half century. Cooke (2012) has advanced the hypothesis that the proliferation of modern information and communication technologies (ICT) are at least partially responsible as enhanced connectivity has reduced the importance of the workplace as a determinant of geographic placement. This thesis provides further evidence in support of the Cooke hypothesis. Specifically, while individuals historically ha ve a higher migratory propensity if they work in occupations seen to be more ICT intensive the work type variable, in relative terms, has reduced significantly over the course of the 44 year study. Additionally, over this period, economically advantaged i ndividuals with greater access to ICT have experienced a decrease in relative migratory propensity. Dr. Richard Coe Division of Social Sciences
1 Chapter 1: Introduction The image of the United States as a migrant society is deep rooted. In some sense, American history is a history of the migrant, who is willing, at times, to make tremendous sacrifices in order to improve her or his lot and strive for the American dream. Whether the image is manifested by the pilgrims that fled Europe in pursuit of religious freedom, the frontiersmen that moved westward to seek profit or cheap la nd, or the migration of African Americans to the industrial north and west following the turn of the twentieth century, the theme of migration in America is firmly ingrained in its conscience. In fact, this idea is so pervasive that it has contributed to a recent history of American media and work in the social sciences that inaccurately portrays the country as being ever increasingl y mobile. 1 Fischer (2002) refers to this phenomenon as being 1 Wade Clark Roof and William Mckinney moving that Vance Packard some years ago entitled a book on the subject A Nation of Strangers Such movement takes a heavy toll on social relationships. It results in weakened ties to family and kin, to neighborhood and community Robert Wuthnow (1994, pp. 5, 22) wrote Restless Nation Jasper states that current mobility is the same as earlier generation s (Jasper 2000, p. 71). 2001).
2 Rather, the United States over the past half century has developed a trend of ever decreasing mobility. There have been periods of increased mobility within the business cycles that compose the past fifty odd years (migration is pro cyclical), yet the overarching trend indicates that the era of a hypermobile Am erica is swiftly receding. If not the economy, what is contributing to this prolonged trend? There is some evidence that the transition of the aging baby boomer cohort into a less mobile demographic has contributed to the aggregate reduction. However, beca use there are observed mobility reductions in nearly every demographic category, aging baby boomers, along with any other demographic factor, is simply not enough to explain the trend. According to Wolf and Longino (2005), Cooke (2011, 201 2 ) and Molloy, Smith, and Wozniak (2011), non economic and non earner coup les, increased indebtedness, an equalization of place specific qualities (i.e. the introduction of air conditioning into Southern and Southwestern climates), and the proliferation of technologies that enhance connectedness and serve as a substitute for tra ditional migration. The purpose of this thesis is to examine, specifically, the impact of advances and adoption of information and communication technology (ICT) on migration behavior in the United States. It is hypothesized that different modern communi cation technologies have contributed and will continue to contribute to an aggregate reduction in domestic
3 migration rates. For example, modern air travel has made it possible for workers to commute across regional boundaries, while videoconferencing, emai l, and other advances in digital connectivity have increased the possibility of working from home for many white collar workers in the United States. Unfortunately, there is no data currently available to perfectly test for technological use among migran ts. In a similar study, Cooke (201 2 ) attempted to use per capita access to telephone lines (the only ICT related statistic offered by the U S Census Bureau), yet encountered trouble due to the near universality of telephone access in the contemporary Unit ed States. To remedy this problem, a proxy variable is created for this thesis that separates individuals by the amount of ICT use common to their occupations. The variable is a dummy, designating an individual as either generally working with or without ICT in its line of work. Because of the unreliability of both the creation and implementation of the proxy variable, the hypothesis ultimately cannot be accepted or rejected. However, the results demonstrate strong further evidence supporting the hypothe sis that ICT proliferation negatively affects the aggregate propensity to migrate. It is found that after controlling for age, education, and macroeconomic factors, since 1968, migration rates of individuals in ICT intensive occupations have declined at a significantly faster rate than those working in ICT non intensive occupations. Additionally, it is shown that those with greater access to modern technology (the eco nomically advantaged young and educated) have also experienced a greater relative reduction in migration propensity over the 44 year period. In light of the evidence in support of the hypothesis, t his thesis argues that the emerging proliferation of ICT i nduced connectivity constitute s a new form of mobility
4 that cannot be quantified using traditional definitions of migration. If it is accepted that is in The implications inherent in this thesis are of sure value to policy makers. Most importantly, given that ICT proliferation will almost certainly increase exponentially, an aggregate reduct ion of domestic migration rates should persist into the foreseeable future. Understanding and being able to predict migration trends is essential for efficient resource allocation in both the public and private sectors. In order to investigate the accuracy of the hypothesis, this thesis will begin Chapter 2 with a background on migration literature, examining methods, theory, and empirical information. Following will be a survey of ICT literature with an emphasis on links between ICT and mobility. With grou nding in the literature, Chapter 3 will survey the available data and defend the methods undertaken to empirically test the hypothesis. Chapter 4 will report the findings of the empirical study. Lastly, Chapter 5 will offer concluding remarks, presenting p otential policy implications, in addition to avenues for further research.
5 Chapter 2: Literature Review Overview of Migration Studies While the nature of migration has become more complicated as time has progressed, traditional migration theory makes distinctions between international, regional, and local migration. According to W.A.V. Clark (1986, p.11) have been connected to labor and housing markets, regional moves have been related to job behavior, and international moves have b een related to p olitical and religious These principles can still be used as a rough general framework; however, the patterns of migration separated by scale are increasingly less predictable. For example, as Clark points out, regional moves in d eveloped nations for career based reasons became a less significant portion of total regional movement during the second half of the twentieth century (Clark 1986 ). In 1998, the Census Bureau began including the topic Reason for Move in its annual Current Population Survey (CPS). Comparing the CPS results from 2009 with 1998 shows the trend of reduced mobility for j ob related reasons to continue. This section will explore the intricacies of each type of mobility and migration.
6 Geographic Units to Define M igration Differing definitions As d efined by the United States Census, domestic migration Bureau). However, the term migration, as used in the literat ure, often means the change of residence by an individual that removes him or her from a particular labor market, in favor of another (Clark 1986 Wolf 2005 pg6, Molloy et al 2011 p. 3). There are multiple ways that a demographer may choose to classify reg ions for migration purposes. One common method is the use of metropolitan areas, which are geographic areas defined by their ability to share infrastructure, employment, and other resources. Although the definitions vary slightly, metropolitan statistical area (MSA), core based statistical area (CBSA), and Economic Area (EA) are all names used to define metropolitan areas. Some demographers prefer to use geographical areas defined by governmental boundaries, however. Both methods have unique advantages and disadvantages. Molloy, Smith and Wozniak (2011) pinpoint three main disadvantages to the use of metropolitan areas as basis to define migration. First, metropolitan areas only include areas with relatively high density. Thus, a relocation either to or from a rural area will not be included in metropolitan area based migration statistics. To remedy this problem, PUMAs, or public use microdata areas could be used instead of MSAs. These areas are similar in concept to MSAs in that they frequently incorporate multiple counties together, yet PUMAs include all of the United States. PUMAs do not describe labor markets as well as MSAs, however, and are not frequently used to study migration. The second
7 disadvantage of metropolitan areas is that they are defined by factors that change, such as population, housing markets, and employment. Because of this, issues may arise regarding inconsistency when studies are done to measure changes over time. Third, metropolitan areas are not included as labels in many data sets ( Molloy et al 2011, p. 4). Using state or county boundaries solves some of the problems related to the use of metropolitan areas. Not only does this method include all of the country, but the boundaries do not change over time. However, in part because thes e boundaries do not change, the area they represent is rather arbitrary. For instance, many labor markets, such as New York, extend beyond state lines. Someone could change residence from Manhattan to Hoboken, while still working the same job. Similarly, a move from Washington Heights to the Bronx would represent a move across county lines, while the individual would remain in a very similar labor and housing market. Time Period to Define Migration The second decision a researcher must choose to make when studying migration is the time period with which to judge it. The shorter the time period, the more accurate and useful the migration statistic is to the researcher. Migration is reported as an individual having a different residence than the last time obs erved. If an individual has moved five times in five years, a migration study done with five year intervals would only record the individual as having moved once. A study done with one year intervals would have a higher likelihood of recording an appropria te number of observations. In addition, if a greater time period is used, there is a greater chance that an individual or household would move back to the original location, resulting in no observed migration.
8 This problem is especially relevant when study ing life time migration, since many individuals choose to move back to their birthplace (state, county, or MSA) after a life spent with other changes in address. Migration data will be discussed further in the empirical section of this thesis. Forces of M igration Correlates Age The demographic attribute that indicates the highest likelihood of adult migration is age (Clark 1984, Long 1988, Wolf 2005, Cooke 2011). Generally speaking, migration rates peak in the 24 34 age range, and recede consistently as the age of individuals increase s When migration rates are charted against age, there is sometimes a modest local maximum around retirement age. However, it is most often that the only maximum is the global maximum, which occurs around the peak migration period of 24 34 years (Long 1988). Years in which a non global local maximum due to retirement may h ave existed are 1949 and 1971 (Wolf 2005). The reasons for this phenomenon are fairly obvious. As an individual is in the stage of young adulthood, she or he is beginning her or his adult life, perhaps leaving college, or otherwise entering into a career. The individual is less established, and faces fewer costs to migration. In addition, the human capital approach (which will be outlined more in depth later on) suggests that as an individual is younger, a move has more
9 s life is ahead, there is more time to accumulate bene fits of relocation (whether such are income related or related to other benefits) (Long 1988). Migration of children follows a different trend, however. For children, or the life period before the pea k in the mid twenties age range, migration is at its greatest height during the infant stage, and gradually decreases until high school. Then, following high school, migration rates increase until they reach their eventual global max in the mid twenties (C lark 198 6 ). The reason for this is that households become more significantly invested in a particular location as a child ages and becomes established. In addition, an aging child relates to an aging parent, which is a correlate of reduced migration. Education/Wage The second major correlate of migration is education. According to Clark (198 6 ), an individual or household with a college degree is five times more likely to migrate than an individual with only an elementary education, and almost twice as likely to migrate as an individual with only a high school degree. Bartel (1979) also demonstrates that education has an independent effect on the decision to migrate. Bartel (1979) says that been explained as being propensity to migrate, wages may not be ( Bartel 1979, Long 1988). Wages affect migration differently depending on whether or not migration is related to job separation.
10 In most scenarios, wage and migration are positively related. However, in the instance that migra tion is accompanied by job separation, wage and migration are negatively related (Bartel 1979). The explanation for this is the negative effect on wage that is produced by job separation. Regarding occupation, Bartel (1979) argues that because jobs requir ing higher education more frequently operate in national, rather than regional labor markets, individuals in such sectors would be more likely to migrate. Additionally, job tenure is a reliably negative correlate to migration. As job tenure increases, job separation becomes less likely, in part because of the strong relationship between tenure and job specific training. Economic Forces Labor Market Conditions Ceteris Paribus, unemployed individuals, or those dissatisfied with their jobs are more likely to migrate than happily employed ones (DaVanzo 1978). Furthermore, when analyzing the decline in migration that occurred between the years 1999 and 2009, Cooke (2011) attributes an increase in migration of 5.67% of the total decline to increased duration o f unemployment. Because unemployed individuals are more likely to move, the migration rate during this period was 5.67% higher than it would have been otherwise. Interestingly, the study also finds that the labor market of a region serves more as a force p ulling migrants towards the location, than pushing them.
11 Additionally individuals who relocate for job related reasons, but fail to find a suitable job after the move employment related, indiv iduals who are dissatisfied with their most previous move According to Sasser (2010), relative labor market conditions is a significant determinant of migration. If labor market conditions in are less satisfactory than another location, the individual has a greater likelihood of leaving the current location. However, while employment opportunities can impact an nce of the variable has fluctuated, historically. The importance of labor market conditions, which is measured by unemployment insurance (UI) claims in the study, doubled between the first (1977 1986) and second (1987 1996) decade of the study, yet returne d to its prior level of significance during the third decade (1997 standard deviation deterioration in the relative UI claims rate simi lar to that which occurs during a recession inc reases the outflow of migran ts from the origin to the destination state by Incomes It has already been mentioned tha t wages affect migration differently, depending wage suggests higher migratory potential, yet an unemployed person (who has a low wage) is also likely to move (Bartel 1 979). In addition, the unemployed are generally
12 more responsive to other economic determinants of migration (family income, origin wage rates, and expected earnings increases) than those satisfied with their job (DeVanzo 1978). Income effects are seen by D eVanzo to be strong and negative for the unemployed. Unless the income bonus is conditional upon its use for moving purposes, the individual is more likely to use the bonus to ease the burden of unemployment, and remain in its original location. However, a s Mueser and Graves (1995) point out, there are other factors that make estimating the determination of wages on migration difficult. Multiple reasons are presented. First, a researcher must correct for worker quality, which is difficult to do. In addition wages must be examined over time to study migratory trends, which leads to problems since wage changes are particularly difficult to analyze in the short run. The reason for this is that employers often adjust real wages by altering employment standards, rather than adjusting nominal pay. Lastly, wages are often a function of land rents at a particular location. According to the authors, researchers have often measures of em Relative per capita incomes for regions is also an important component of income based migratory force, historically, and still is today (yet to a diminished degree). Individuals, all else held constant, will choose to migrate to lo cations with better per migration has diminished considerably since the late 1970s...A one standard deviation increase in relative per capita incomes reduced the numb er of out migrants by roughly 20 percent in the 1977
13 Housing Market Conditions Compared to other economic factors that influence migratory behavior between regions, housing market conditions have risen in importance. For instance, during the 1977 1986 decade, a one standard deviation increase in relative housing affordability had no significant impact on out migration, yet in the two decades since, the same increase responds with a 4% decrease in the out migration rate (DaVanzo 1978). However, similar to their critiques of wage as a determinant of migration flows, Mueser and Graves (1995) argue that home prices may be indicative of quality, which is difficult to correct for. Accordi ng to Zabel (2009), a renter has a higher likelihood of moving than a homeowner. In fact, in 2006, the US Census Bureau released information indicating that in the same year, 30 percent of all people living in renter occupied housing units lived elsewhere a year earlier, while the moving rate of residents of owner occupied housing units was 7 percent (Census Renters). The reasons Zabel attributes to a higher migratory i As an abnormally large number of Americans have seen their real estate values drop below the figures bound to in their mortgages, there is an obvious resistance among those Ferreira, Gyourko, and Tracy (2008) show in their study that when introduced, negative equity lowers the two year mobility rate by 4 percent age points ceteris paribus. For perspective, the baseline two year mobili ty rate used in the study was
14 11.4%. This indicates that negative equity should lower mobility by more than a third, which has strong implications. Additionally, there is speculation that a declining housing market hurts itself by limiting the confidence of buyers and renters. If the future value of property is perceived to be low er than it is currently valued renters are not likely to buy homes, and homeowners are not likely to move to more desirable homes (Molloy, Smith, Wozniak 2011); And unfortunate ly, this trend does not correct itself by a poor market signifying potential for growth in value, because homeowners are traditionally very unlikely to list their home for less than they owe on the mortgage, regardless of what the current appraisal indicat es. prospects, there is a very literal way homeownership and the housing market collapse has As homeowners are occasionally e ventually forced to foreclose on their homes, they are similarly forced to move (Ferreira, et al 2008). Human Capital Model of Migration neteenth century (Grigg 1977). The works have been thousands of migration studi es in the meantime, few additional generalizations have been
15 Ravenstein was the first to put forth the idea that individuals choose to move because of economic reasons, which served as a launching point for human capital theories of migration i n the twentieth century. true during the time of publication (Grigg 1977). economic advantages, chiefly, differences in wages, are the main causes of migration (Morrison et al 2010), and the notion that individuals make decisions to migrate for the purpose of general self improvement was first discussed by Larry A. Sjaastad (1962). According to Sjaastad (1962), as individuals relocate to make themselves better off, allocation. Migration can then be viewed as an investment in human capital. The investment cost is the costs imposed by the migration, and it is imposed immediately. This can include the economic costs of movement, such as buying a new home or renting a moving truck, or less tangible costs such as losing proximity to friends. However, to Sjaastad, utility gained from migration (and the costs imposed) was still predominantly related to economics. The decision to move would imply that the long term economic benefits fro m a human capital investment (in the labor market) are determined to outweigh the initial costs. Clark (1986, p.67) acknowledges two main advantages to the human capital approach. First, the model considers that benefits from migration do not necessarily occur instantly, but over a period of time. The reason that this is valuable is that it helps explain why migration rates trend downward with age; as an individual ages, she or he also has
16 less time to accrue benefits from migration. In addition, a person may move without immediate expectations of benefit. Secondly, the human capital model can (although at times did not) incorporate non economic forces. While many models composed by those working in economics only consider monetary costs and benefits, the h uman capital approach considers proximity to cultural facilities, proximity to friends and relatives, climate, access to public goods, among other non economic assets. The way in which the human capital model is used by the researcher varies. The human cap ital model, as introduced by Sjaasted, views migration as increasing an similarly under stood migration decisions to follow a cost/benefit analysis, yet expanded for movement. Khwaja (2000) uses geographic wage differentials to model migration patterns and Kan (1998) also treats mobility as an investment in employment. In these models, which are fundamentally human capital models, net discounted returns expected from moving are estimated to exceed returns from not moving. Morrison et al (2010) mention s the paradox between the focus on migration as an enhancement to returns to labor, and the empirical data indicating that moving for job related reasons account for 30 percent of migration between labor markets. According to st people of working age, employment remains a necessary movement as primarily driven by consumption based forces (location), associated lifestyle, and family and social reasons.
17 Reasons for Moves Ignoring forced moves, Clark (1986, p.41 45) classifies reasons for moves into two main categories: adjustment and induced. 1986, p.41 42). Reasons for adjustment moves include: housing (space, quality/design, cost, and tenure change), neighborhood (quality, physi cal environment, social composition, and public services), and accessibility (workplace, shopping/school, family/friends). Reasons for induced moves include: employment (job change, retirement) and life cycle (household formation, change in marital status, and change in household size). It should be noted, however, that these classifications are not necessarily exclusive. For example, a move due to accessibility to workplace could occur because of a change in employment, or despite it, even though relocatio n for accessibility reasons is generally classified as an adjustment move. Similarly, an individual could change employment, while choosing to not change its residence. CPS Survey The first surveys conducted of reasons for moving date to the 1946 CPS, w hich
18 moving question was in 1963. This time, it phrased the question differently, in order to put emphasis on the destination, rather than where one was leaving. The 1963 CPS asked 1946, the survey only asked whether or not the individual left the county In 1963, the CPS considered both intra and inter county moves. In the 1970s, the American Housing on individuals. Additionally, the AHS listed far more options for reasons for moving than the previous surveys, including retirement and climate. However, the data collected by these surveys had little value to researchers. As dies [before 1988] of reasons for moving have not produced comparable time series data, for some have been based on non representative samples and others have asked questions that seem biased toward producing anticipated ed towards solving these problems, in particular, focusing on the issues of consistency and frequency. That year represents the first of the annual inclusion of a reasons for moving section in the CPS. Census reports of the reasons for moving section of th e CPS present the survey as choices broken into three main sections: 1) family related reasons, 2) work related reasons, 3) housing options: attend/leave college, change of climate, health put into either family, career, or housing reasons. For example, to attend college is included in work related reasons. The rest is amb iguous. Regarding the main category
19 choices, family related reasons include: change in marital status, to establish own related reasons include: new job/job transfer, to look for work/lost job, closer to work/easier commute, an related reasons include: wanted to own home/not rent. new/better house/apartment, better This categorization includes all reasons for voluntary movement outlined by Clark (1986), h owever the distinction between adjustment and induced moves is not made. In 1998, family related moves accounted for 27% of total movers, work related moves accounted for 17.1% of total movers, and the largest reason for moving was housing related, at In 2009, the statistics were relatively similar. Family related moves reduced to 26.3%, work related moves increased to 17.9%, and housing to 9.8%. Figures 2. 1 and 2. 2 were produced by the author using data available from the US Census website.
20 Figure 2. 1: Figure 2. 2: Data Compiled From http://www.census.gov/hhes/migration/data/cps.html
21 It seems that there may be a trend away from job related movement re lative to other types of migration, which would support the hypothesis that ICT proliferation has contributed to aggregate migration decline via increased connectivity among separated workplaces. However, the data is not entirely conclusive. Not only are t he trends not completely convincing, the data is not available for a long enough period to account for business cycle fluctuations. Migration Trends Since 1968, there has been a decline in internal migration in the US, and since 2001, that decline has been steeper. In 1968, US annual interstate migration rates peaked at 7.0%, and inter county rates experienced a peak at 3.6%. In 2009, these same figures declined to their respective low points at 3.7% and 1.6%. This trend has been accompanied by perio dic spurts in migration the mid 1980s, and late 1990s are examples of these periods. However, these spurts are easily explained by coincidi ng economic expansion. Figure 2. 3 below illustrates the downward trend in US migration rates.
22 Figure 2. 3: Source : Cooke (2011) Interestingly, the intercounty and interstate migration trends only represent a small segment of the total decline in residential mobility in the United States. Whereas the quadratic trend lines mapping the migration rates in Figure 3 show a decline since perhaps 1970, Fischer (2002) shows that when examining the migration trends of all American migrants, there is a steady reduction in rates since the late 1940s. This is because there has been an even steeper and longer reduction in local or intracounty migration compared to intercounty and interstate. Molloy et al (2011) separates reasons for change in aggregate migration rates into three categories. One category involves the consistent relationship between the economy and migration rates. As the economy is healthier, individuals are more prone to migrate. The second category of forces of migratory change is the distribution of individual characteristics that are
2 3 associated wit h benefits in net migration (i.e. demographics). For example, because age is generally negatively related to migration rates, the aging baby boomer population will likely decrease aggregate rates. The third and final explanation for change in migration pa tterns that Molly et al list relates to changing migration behavior for individual groups of people. Because this explanation for migratory change controls for both economic that puzzles migration researchers. Cooke (2011) uses regressions to determine the significance of variables affecting the negative migration rate change between the two years 1999 and 2009. The data used is from the Current Population Survey ( CPS) and the type of move analyzed is intercounty. The results show that 63% of the decline can be attributed to economic factors mostly the economic crisis that began in 2007. A change in the makeup of the population, by age and household structure, was r esponsible for an additional 17% of the migration decline. The last portion, 20% of the decline, represents a more fundamental decline in domestic migration rates th e United States has seen since the 1960s. Migration rates may rise momentarily as the housing and job markets recover, yet will eventually dip again, as historical trends indicate. The secular shift indicates that as the years progress, individuals are co nsistently less likely to change addresses. The reason is unknown to social researchers, yet at least one variable has altered American habits in a fundamental way. Fischer (2002) points out ocial sciences which asserts that with modernity, society becomes less rooted and more mobile. Clearly, in
24 terms of physical relocation, the grand narrative does not hold up next to facts (Fischer 2002, Molloy et al 2011, Cooke 2011). Cooke (201 2 ) presents a few possible explanations for the emerging culture of rootedness in the United States. One theory is that as the role of women in the workplace has improved since the 1960s, so has the prominence of the dual earner couple, which has made relocation tric kier for an increasing amount of people. The logic is that a move would require two individuals to find new jobs, rather than just one. Cooke (2009) cites Shauman (2009) as partial evidence that the rise of the dual earner couple has contributed to a reduc tion in aggregate US migration rates, yet on January 11, 201 2 the article could not be found on the internet. Conversely, the findings in Cooke (2009) indicate that the rise in dual earner couples have had no effect on the migration decline. Possible expl anations supporting the figures in Cooke (2009) include that a dual earner couple may have more economic resources to support a move, and in practice, the female earner (in a traditional male female couple) often follows the man without a job in place at t he destination (Morrison et al 1988, Cooke 2003). Additionally, as the real wage in the US has fallen, Americans have increased their indebtedness to compensate (Cooke 201 2 ). On one hand, this phenomenon has likely increased the proportion of dual earner couples, as families have a growing need for supplemental income. Most importantly, however, increased indebtedness is known to lower migration rates, as shown by Ferreira et al (2008). Moving is a financial expense that more indebted households are unable to afford. This has been most evident during the recent migration slowdown, yet is not limited only to the current period (Frey 2009).
25 Individuals are less able to capitalize on place specific utility differences, because of both the rise in dual earner families and increased indebtedness. According to Greenwood et al (1991), capitalization of place specific utility differences is tied very closely to total migration behavior. A leveling of place specific utility differences could be a cause for US migrat ions rates to decrease, as there is growing uniformity among places; and as locations vary, housing prices and wages are said to be adjusting more appropriately to compensate for utility differences more appropriately than the past. However, according to P artridge et al (2012), there is little connection between a leveling of place specific utility and a decline in migration rates. Finally, Cooke (201 2 ) cites the potential impact technological growth has had on reshaping fundamental migratory behavior. Pre sented as fallacious by some world, holds up if certain modern technology is viewed as a new form of mobility. Like range, new technologies continue to make it easier to change jobs without changing residence. For instance, more comprehensive and affordable air travel has made interstate and interregional commuting possible. In addition, the modern worker has constantly improving means for virtual commuting due to the proliferation of internet access and capabilities.
26 Information and Communication Technology they essentially eliminate the link between cost of communication and the distance 07 ) facilitated (virtual) reality for contemporary society, mobility must be defined by movement of people and ide as. This expands the perhaps too strict definition used by demographers to track spatial change, a no broadly accepted empirical link between the declining rate of domestic migration and the increase in movement in US working life, there is ample theoretical support for the idea that the latter is at least partially responsible for the former. Measur ing ICT by technology adoption presents a rather clear picture of technological growth in the United States. As shown by Figure 2. 4, b y the 1960s, the country had 50,000 telephone lines. Between then and now, growth was rather steady, leading to over 200,0 00 lines in the US today. The growth of private car use since the 1950s has nearly m irrored the growth in telephone lines (although a little flatter). In 1960, there were approximately 60,000 private cars in the US, and by the late 1990s, there were approx imately 150,000. Alternatively, the growth of internet, personal computers, and
27 mobile phone use is concentrated in the past 30 years, and has exhibited a much more exponential growth rate. There has been similar growth in Europe (Ioannides et al 2007). Figure 2.4 : Graph source: Comin and Hobijin (2004) While the source is perhaps not academically credible, the blog Skype Numerology has reported Figure 2. 5, which is intended for use here as an illustration of the exponential growth of webcam use during the past decade.
28 Figure 2. 5: Search Unified Communications (Trost 2010) included Figure 6 in a report pertaining to business adoption of video conferencing. The source used by Search Unified Communications was Nemertes Research. Figure 2. 6 indicates that nearly three quarters of organizations that participated in the Nemertes study were planni ng to deploy some form of video conferencing, with the strongest growth areas being desktop and have increasingly become the norm in work organizations during recent d Videoconferencing is cited as the most technologically advanced course for long distance communication. While room based videoconferencing still remains the favorite videoconferencing method for businesses in the United States, internet based serv ices
29 such as Skype have been gaining tremendous popularity due to low cost and convenience (Julsrud et al 2012). Additionally, Julsrud et al (2012) find that widespread adoption of videoconferencing in business (more so room based than internet based) has appeared to reduce overall business travel. Figure 2. 6: According to Hardill and Green (2003), technology and telecommunication are work...with tasks for paid work undertaken in a variety of locations such as while traveling, whilst at home, and networ business travel and commuting, and business travel and migration is becoming more
30 ambiguous, as transportation networks improve, and employers are increasingly paying for temporary accommodations (hotels or rented apartments) for employees (Hardill 2002). Simply put, the workplace is losing its significance as a determinant of geographic placement, because individuals are less likely to need to focus worktime at a specific location. Additionally, when able to make decisions on where to locate, individuals are increasingly choosing to live in areas that allow for easy access to a range of employment opportunities, such as near major highways rail networks, and airports (Green et al 1999). By locating within re ach to many opportunities, the worker is reducing the significance of her home as an access point to a particular job. ICT advances have certainly also led to the proliferation of devices that have contributed to the reduced importance of the workplace. A ccording to Hardill and Green costly facilities, and equipping employees with portable notebook computers and telecommunications equipment to enable them to work from home, in cars, or in some Some employers are even eliminating dedicated office sp ace as an entitlement, completely. Modern technological advances such as the internet, telephon e and transportation infrastructure enhance connectedness in a way that used to require urban agglomeration (Ioannides et al 2007). Certainly, urban agglomer ation is still an occurring and somewhat necessary reality driven by both economic and social forces. Yet, as people and
31 information can increasingly travel more freely across space, economic activity will similarly become more dispersed. According to Ioan nides et al (2007), populations are located according to both agglomeration and dispersion forces Agglomeration forces pull individuals to dense urban areas, as high population areas lead to human capital and knowledge spillovers, along with advantage s in labor market pooling. Dispersion forces include high rents and pollution that are consequences of highly concentrated populations. As ICT improves, agglomeration forces weaken, dispersion forces increase in relative significance, and populations, alon g with economic activity are said to become less concentrated in space. This theory helps explain the reduced internal migration stream to cities seen in the United States since its peak decades ago. However, it says little about the overall trend away f rom domestic mobility, since the theory indicates that there would rather be movement driven by dispersion forces, away from cities.
32 Chapter 3: Data and Methods Data There are three main data sources used by researchers to study migration rates in the United States: the Internal Revenue Service (IRS) migration data; the decennial U.S. Census, paired with the relatively recent (since 2000) annual supplemental American C ommunity Survey (ACS); and the Annual Social and Economic Supplement of the Current Population Survey (March CPS). Decennial US Census/ACS Survey The Census has been collecting data decennially since 1790, and has been collecting data relevant to migrati on research since 1940. Since 2000, the Census Bureau has incorporated the American Community Survey as a supplement, which collects data
33 information they need to plan investment s and services. Information from the survey generates data that help determine how more than $400 billion in federal and state funds IRS Data (Gross) The IRS migration data are derived from the Individual Master File (IMF) aft er the IMF is sold to the Census Bureau. In order to process tax returns into migration data, the IRS and Census Bureau must first strip the Social Security Number (SSN) and taxpayer name from each return, replacing the information with an identification n umber referred to as a Protective Identification Key (PIK). The Census Bureau then geocodes according to the United States Post Office, then translated into classification b y county, CPS Data The Current Population Survey is sponsored jointly by the U.S. Census Bureau and the U.S. Bureau of Labor Statistics (BLS). It has been conducted monthly since 1940 to measure unemp loyment following the Great Depression, and is the primary source of labor statistics for the population of the United States. Beginning in 1948, the CPS has expanded to include the March supplement, or the Annual Social and Economic Supplement (March CPS) The March CPS has included a question on migration since
34 1963, explicitly, whether or not respondents were living in the same house one year ago (King 2010). Each data source has unique strengths and limitations. The decennial Census and ACS provide m igration data going back as far as 1850, and sample approximately 3,000,000 people per year since the ACS began conducting itself in 2000. Because of its enormous sample size (ACS is 30 times larger than the March CPS and decennial census is 100 times larg er), the ACS is most appropriate for local and state level studies (non national studies needing finer statistics) In addition, the ACS has different opportunities for regional classification than the March CPS or IRS data do. The March CPS classifies an nual migration as intracounty, intercounty, interstate, or international, and migration over a 5 year period as all plus migration to non contiguous states. IRS migration data is defined only by intercounty moves. Alternatively, the ACS is more flexible, e xpanding definitions of migration to include localities, state economic areas (single counties of groups of contiguous counties within in the same state that had similar economic characteristics in 1950) congressional districts, and metro areas. However, A CS does not have a county based definition of migration. The IRS, along with the March CPS, provide annual state and county based definitions of migration that extend well beyond the hardly decade old ACS. The IRS has calculated inter state migration rate s since 1975 and intercounty rates since the early 1980s. Migration rates using CPS microdata go back to 1963. Because of their ability to
35 conduct time series spanning multiple business cycles, migration data from the March CPS or IRS are especially useful to researchers studying historical trends. Due to their method of retrieval, the IRS and March CPS differ starkly with regards to certain strengths and limitations, however. According to the IRS, because their data are obtained from income tax records, t inclusivity may actually be an issue to be wary of since a very significant portion of the United States file no income taxes, thus any project based on IRS data will surely under represent the poor and elderly. I n 2010, 45 percent of households owed no federal income tax (Altshuler and Williams 2010). The March CPS also is a bit misrepresentative of the US population, since beginning in 1976, an oversample of Hispanics (approximately twice as many as proportional there are no weights provided by the IRS to correct for its undersampling of the poor and elderly. Additionally, IRS data is available only to purchase. For this thesis, the migration data from the March CPS is the most appropriate. It may not have as varied of ways to define migra tion as the decennial Census and ACS (the significance of which is reviewed in the literature review chapter), but its intracounty, intercounty, interstate, and abroad definitions are suitable enough for this project. Alternatively, unlike the data from th e decennial Census and ACS, the March CPS allows for a time series that spans multiple business cycles, which is necessary for this project. It
36 is because of these reasons that the March CPS is more suitable than data obtained from the decennial Census and ACS. Compared to the IRS data, the March CPS is most significantly different with regards to its sample. While both samples are not perfectly representative of the US population, data sources for the March CPS provide a weight to correct for the inaccurat e sampling, while there is no known published weight associated with the dissemination of IRS migration data. Additionally, March CPS migration data extends twelve years beyond the first year of IRS migration records. Lastly, March CPS data is more availab le in public use data sets. IPUMS CPS The March CPS data are sourced from the Integrated Public Use Microdata Series (IPUMS) (King 2010) and the CPS version of IPUMS is referred to as IPUMS CPS. From the IPUMS CPS web site: IPUMS CPS is an integrated set of data from 50 years (1962 2011) of the March Current Population Survey (CPS)....To make cross time comparisons using the March CPS data more feasible, variables in IPUMS CPS are coded identically or "harmonized" for 1962 to 2011...IPUMS is not a collection of compiled statistics; it is composed of microdata. Each record is a person, with all characteristics numerically coded. In most samples persons are organized into households,
37 making it possible to study the characteristi cs of people in the context of their families or other co residents. Model: In order to determine the influence of ICT growth and proliferation on yearly migration rates, this study estimates logistic time series regressions from every year available be tween 1968 and 2012, inclusive. There are 29 years where data is available for all variables needed for the study. These years are: 1968 1971, 1976, 1981 1984, 1986 1994, 1996, 1998, 2000, 2002, 2004, 2006, 2008 2012. Specifically, the logit of the an nual intercounty/interstate migration rate (p) among all civilians designated as heads of household is estimated through Maximum Likelihood as a function of a vector of independent variables : Variables Dependent Variable: Migration Status The Dependent Variable chosen for the study has been developed from the IPUMS CPS migration variable MIGRATE1, which indicates whether or not a respondent has changed its residence since the previous March CPS survey. Other
38 migration vari ables provided by IPUMS CPS include a variable that indicates that an individual has moved within a five year period, and a variable that indicates that an individual has moved at least once since 1975. As previously mentioned, the latter two methods to de fine migration are potentially more misleading, since an individual could move countless times within the specified period, yet would only be counted once as a migrant. The one year statistic is the shortest range available to define a migrant, thus it pro vides the most accurate representation of aggregate migration tendencies in the United States, and is used in this thesis. Respondents that reported living in the same house as the previous year were asked no further migration related questions, yet indiv iduals that were identified as migrants were asked about the city, county, state and/or US territory or foreign country where they resided one year ago. This allows the variable to be subdivided into sets by type of migration: those that have moved within the same county; those that have moved between counties, but stayed within the same state; those that moved between states; and those that have migrated from outside the United States. For the purposes of this thesis, the dependent variable was created as a dummy have moved between counties but not between states, and individuals that ha ve moved between states. Individuals that have moved from abroad were eliminated from the study completely, and individuals that moved only within a particular county were grouped in By categorizing local intracounty movers as non migrants, the migration rates found in this study will be much lower than commonly reported
39 migration figures. Similarly, the decline in migration found will be less severe than the aggregate decline, since the decline of local migration has been more significant than nonlocal migration over the years analyzed (Fischer 2002). The reasoning behind defining the dependent variable in such a way relates to the type of movement that ICT proliferation is seen to impact: movement for work related reasons. Fur significant among moves between counties. Intracounty movement is relatively much more strongly associated with movement for housing and family reasons, and would have little signific ance in determining the effect that ICT proliferation on migration patterns (Clark 1984) Additionally, movement from abroad is often heavily influenced by more complex geopolitical factors, and was removed from this analysis for that reason. All person records are assigned a serial number that indicates the household the individual identifies with. Assuming the household the individual identifies with does not change, the serial number remains the same as individuals complete surveys over multiple years. In order to uniquely identify all CPS data, information indicating the year and month of the sample are also included. Because the data relevant to th is study is yearly and all pooled from the March CPS, the figure identifying the data as March was eliminated from the set. In order to deal with the complicating factor that individuals often move to follow a member of their household (whether parent, sp ouse, or other), and that their reason for moving would thus have nothing to do with migration impulses that this study is interested in, individuals were only used in the analysis if they identified themselves as
40 March CPS survey collects detailed data on the relationship of each household member to the head, thus all non heads can be easily filtered out. Independent Variables Age As previously stated, the demographic attribute that indicates the highest likelihood of adult migration is age (Clark 1984, Long 1988, Wolf 2005, Cooke 2011). The ages observed in the survey either range from 0 to 99+ (reported as 99), or 0 to 90+ (reported as 90) between the years 1988 2002. In order to make the data in each pe riod compatible, individuals aged 90 or older were removed from the set. After adjusting the data to include only individuals that identified themselves as heads of household, the age minimum became 14. Thus the age range of individuals studied is 14 89. The variable was then turned into a series of six dummy variables divided into the following age categories: 1) 14 29 2) 30 39 3) 40 49 4) 50 59 5) 60 69 6) 70 89
41 The following table lists the mean value of the dependent variable for each age category. T able 3.1 Age Mean of Dependent Variable (Migration Rate Mean) 1968 1968 1978 1979 1989 1990 2000 2001 2012 2012 14 29 16.8835 15.7425 14.5212 14.4152 10.7834 9.8246 30 39 7.4424 6.7840 6.7738 6.5061 4.4649 4.2291 40 49 3.9052 3.6790 4.3206 4.2105 2.5594 2.6437 50 59 2.1753 2.2507 2.9437 2.9896 2.0587 1.8496 60 69 1.6596 1.8906 2.2758 2.0663 1.6405 1.2802 70 89 1.0204 1.1394 0.9773 1.4681 1.2184 1.4740 Educational Attainment The variable used for the educational attainment variable is the combination of two IPUMS provided variables which measure educational attainment in different ways. high est grade of school or year of college completed. The second education variable, which covers 1992 and the years that follow, classifies individuals by the highest degree or diploma obtained. To make the two definitions compatible, 4 years of college was c considered equivalent to a graduate degree. A series of 5 dummy variables was made combining the two IPUMS variables. Each of the five dummy variables represents a different le vel of educational attainment. 1) No High School Diploma 2) High School Diploma
42 3) Between 1 3 years of college 5) 5 or more years of college or a graduate degree The following table lists the mean value of the dependent variable for each educational attainment category. Table 3.2 Education Attainment Level Mean of Dependent Variable (Migration Rate Mean) 1968 1968 1978 1979 1989 1990 2000 2001 2012 2012 No High S chool Diploma 4.7 4.678 5.31 4.705 3.546 3.2 High S chool Diploma 7.095 8.308 7.824 5.610 3.635 3.352 1 3 Years of College and no Bachelor's Degree 8.738 8.064 7.691 6.677 4.069 3.591 4 Years of College or Bachelor's Degree 10.58 10.67 9.524 8.361 4.607 4.046 5+ Years of College or Bachelor's Degree 12.04 10.08 8.252 6.871 4.151 4.142 Occupation To test for ICT use as an influence on migratory behavior, an occupation dummy variable was created from an IPUMS provided occupation variable and identifies an individual as either working in a technologically intensive (white collar) occupation, or not. The IPUMS provided occupation variable that was chosen recodes all data in terms of the 1950 Census Bureau occupational classification. Doing so allows for better compatibility for the time series study, as there is a consistent set of occupational codes.
43 The 1950 Census Bureau occupational classification system includes hundreds of occupations divided into the following 12 categories. The categories were then classified further for this thesis as either mostly ICT intensive/white collar, mostly non ICT in workers did not fit clearly into the ICT/non ICT dichotomy, individuals identified by the Census Bureau occupational classification as sales workers were removed from the s tudy. Hence, the variable was divided as a dummy: ICT Intensive = 1, Non ICT Intensive = 0. ICT intensive/White Collar 1) Professional, Technical 2) Professors and Instructors 4) Managers, Officials, and Proprietors 5) Clerical and Kindred Non ICT intensive/Blue Collar 3) Farmers 7) Craftsmen 8) Operatives 9) Operative and Kindred Workers 10) Service Workers (Not Household) 11) Farm Laborers
44 12) Laborers Other 6) Sales Workers The following table lists the mean value of the dependent variable for both occupation type categories. Table 3.3 Occupation Type Mean of Dependent Variable (Migration Rate Mean) 1968 1968 1978 1979 1989 1990 2000 2001 2012 2012 ICT 7.9220 15.7425 14.5212 14.4152 10.7834 3.7656 Non ICT 5.1707 6.7841 6.7738 6.5061 4.4649 3.5116 Macroeconomic Controls The macroeconomic controls used in the regression include the unemployment rate, the labor force participation rate, and the percent change in GDP. The variables were developed using data from Federal Reserve Economic Data (FRED) provided by the St Louis F ederal Reserve. The macroeconomic data for each year is averaged to produce one value to assign to the entire year. For percent change in GDP, the value is an average of all observations within the year of the percent change in GDP from a year prior. For t he unemployment rate and labor force participation rate, the value assigned to each observation is the average value (in percent) for each year. All data is seasonally adjusted.
45 Because the macroeconomic controls had only one value for each year in the study, they were not able to be used in regressions run by year. However, when analyzing the impact on the dependent variable over multiple years, the macroeconomic controls could b e used. Following is the average value of the dependent variable (migration rate) for select values of the macroeconomic controls. Table 3.4 Macroeconomic Controls Mean of DV (all years) Unemployment Rate < 6% 5.7683638 Unemployment Rate 6% or > 6% 5.665533 Labor Force Participation Rate < 65.5% 5.5495414 Labor F orce Participation Rate > 65.5% 6.288575 Change in GDP < 3% 5.00105 Change in GDP > 3% 6.203365 Variables Not Used Because thi s study was conducted with over 1.2 million observations, several variables that are known to impact migratory behavior were omitted in effort to make the data analysis process more manageable. The most important variables were included, and the rest that the literature review identifies as less important were n ot used. These include: Wage/Income, job tenure, residential tenure, housing market variables, climate, public goods, and proximity to friends and relatives.
46 Weight groups of people who are not included in the production of published labor force statistics: (1) members of the armed services, and (2) members of the Hispanic oversample who were n ot interviews in months other than March. [The weight] is based on the inverse probability of selection into the sample and adjustments for the following factors: failure to obtain an interview; sampling with large sample units; the known distribution of the entire population according to age, sex, and race; over sampling of Hispanic persons; to give husbands and wives the same weight; and an additional step to provide consistency with labor force estimates from the basic
47 Chapter 4: Results Historically speaking, individuals employed in ICT intensive occupations have a higher propensity to change residence. In 1968, white collar workers were nearly fifty percent more likely to move than blue collar workers (Figure 4.1 ). Ho wever, by today there is no clear distinction between the migration behaviors of the two occupation types. Both rates follow the pattern of long term aggregate decline, yet migration rates for workers in ICT intensive employment have declined at a rate rou ghly twice as steep. Figure 4.1 0 1 2 3 4 5 6 7 8 9 10 1960 1970 1980 1990 2000 2010 2020 Migration Rate (In Percent) Year Migration Rates by Occupation Type Interstate + Intercounty White Collar Blue Collar
48 When controlling for age and education, the two most influential determinants of migration, the occupation type dummy variable produced parameter estimates that validated the intuition presented by the graph of occupation type migration rates (Figure 4.2 ). Between 1968 and 2012, occupation type has progressively become a less significant determinant of migration. An exponentiated parameter, or odds ratio of 1.188 (which is the value estimated by the logistic regr ession for 1968) indicates that an individual in white collar employment was 18.8% more likely to move after controlling for age and education. 2 In 2012 the occupation type variable had an odds ratio of 0.999, indicating it had virtually no impact on the d ependent variable. Figure 4.2 : 2 Freund and Little (20 00, pg. 179) provide the following information on the odds ratio: is the change in the event odds, defined as for a unit increase in the dependent variable. In [the example in the boo Maximum Likelihood Estimate is 0.0303 and] the odds ratio is = 1.031; hence the odds [the dummy 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 1968 1975 1982 1989 1996 2003 2010 Odds Ratio Point Estimates Year Occupation Dummy Odds Ratios Probability Against Occupation = ICT non intensive ICT-intensive/ White Collar
49 In comparison, some of the parameter estimates for age and education based variables remained relatively stable over the course of the study (Figures 4.3 & 4.4 ). There was minimal variation that contradicted the literature For example, through the determinant of migration, while during all other periods this was not the case. For the most part, however, the quadratic trends of all age and education dummy variables remained relatively flat, and the order of significance mimicked the literature. However, if there is a conclusion to draw from the slight changes of the education and age odds ratio values, it is that the migration implications tied to the attributes associated with the traditionally economically marginalized (the elderly and those with less formal education) have not trended towards insignificance like their economically advantaged counterparts. In fact, in relation to the 14 29 age category, the 70 89 dummy saw its odds ratio roughly double during the years observed, which indicates a relative decline in the migratory edge shared by household heads under 70 years old. Similarly, ree dummies experienced a noticeable decline since 1968, when compared to the no high school diploma variable. This indicates that over the years studied, highly educated individuals saw their migratory edge diminish relative to those without a degree. By considering that economically advantaged individuals have greater access to modern technology, the change in the significance in the age and education dummy variables provide further support for the hypothesis that the proliferation of modern technology h as contributed to a decline in aggregate migration rates.
50 The macroeconomic controls were only able to be used in the regression that spanned all years. The reason they could not be used to analyze variables year to year is because each macroeconomic cont rol has only one value for each year. However, when used in the regression spanning all years, they helped validate the findings from the year to year regressions. Perhaps in part because of the enormous sample size used in the study (over 1.2 million observations), every variable was deemed significant in every instance. Figure 4.3 : 0 0.5 1 1.5 2 2.5 1968 1975 1982 1989 1996 2003 2010 Odds Ratio Point Estimates Year Education Dummies Odds Ratios Probability Against Education = No HS Diploma 5 or more years of college or a graduate degree 4 Years of College or Bachelors Degree 1-3 Years of College HS Diploma
51 Figure 4.4 : The hypothesis that ICT proliferation has decreased aggregate long term mobility rates is supported by the steeper decrease of white collar migration rates relative to the blue collar rate. The yearly trend of odds ratio point estimates of the occupation t ype variable confirms the significance of the picture presented by the migration rate trends. Additionally, the hypothesis was supported by the findings that between the surveye d years, economically advantaged individuals that have greater access to infor mation and communication technology saw a reduction in relative migratory propensity. 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 1968 1975 1982 1989 1996 2003 2010 Odds Ratio Point Estimates Year AGE Dummies Odds Ratios Probability against Age = 14 29 30-39 40-49 50-59 60-69 70-89
52 Chapter 5: Conclusion In order to illustrate the theoretical basis inherent in the hypothesis, and to provide sufficient context with which the reader can fully understand and appreciate the empirical process, the second chapter of this thesis was dedicated as a literature review. Attempts were made to produce connections between the migration and ICT related subsections of the literature review. The available lit erature did not provide for substantial relationships between migration and ICT theory, yet is sufficient as a background for the empirical portion of the thesis. The third chapter describes the data available and the data chosen for the analysis. Additio nally, the variables created are defended in relation to the literature, and a model is presented that tests the impact of occupation type (ICT proxy), age, educational attainment, unemployment, labor force participation, and GDP growth on the dependent va riable created to represent non local migration. There are 29 years included in the study, the earliest being 1968 and the latest 2012, and there are 1 246 280 observations.
53 Finally, the forth chapter supports the hypothesis by demonstrating a steeper dec rease in white collar migration rates relative to the blue collar rate, even after controlling for other factors. The hypothesis is additionally supported by the findings that between the surveyed years, economically advantaged individuals that have greate r access to ICT saw a reduction in relative migratory propensity. Like many academic fields, demographic, and in particular, migration studies lives and dies by the data made available to its researchers. It is this fact that perhaps most appropriately co ntextualizes this thesis. On one hand, migration studies provide to its researchers a vast realm of investigative possibilities; however, the degree to which limited data availability negatively impacts the realization of these possibilities is rather subs tantial. With that said, the idea for this thesis developed in response to a question posed by several papers written in the past few years: What, precisely, are the non economic and non demographic factors contributing to a long run aggregate decline in domestic mobility? Most likely, there are a few possible answers to that question, as evidenced in Molloy, Smith and Wozniak (2011) and Cooke (2012 ). However, the evidence supporting the impact of one variable in particular, Information and Communication Technology on decreasing US aggregate migration rates was rather scarce. Like studies before it, this thesis does not successfully confirm the hypot hesis that ICT have contributed to the decline. However, it is the first comprehensive consolidation of migration and ICT literature for the purpose of supporting the hypothesis. Additionally, with the use of proxy variables, this thesis provides further e vidence in support of the hypothesis. Lastly,
54 not simply a factor contributing to the decline of traditional mobility, but is mobility the same, in an alternative form. For these reasons, this thesis should be considered as a contribution to the literature. In order to make further advances studying this topic, different data must be made available. While data pertaining to ICT adoption and use patterns is abundant there is not much, if any, that can be seamlessly incorporated into this type of individual level migration study. Perhaps a study linking ICT and migration could be done using aggregate ICT data. However, a tremendous amount of in formation made availabl e by Census or IRS provided individual level reco rds would go to waste. The best case scenario would likely entail an expansion of the March CPS or ACS survey to include a wider range of questions, yet other social surveys, such as the NORC General Social Survey (GSS), conducted by the University of Chicago hint at similar potential. GSS already includes migration, as well as ICT questions pertaining to email usage Unfortunately, however, the years in which NORC GSS data are available are scarce, making even a univariate time series st udy difficult. As time progresses and the need for such available data becomes more apparent, it is crucial that greater resources be allocated to allow more complicated studies to be conducted. This thesis and studies befo re it may have supported the idea that there is a relationship between technology and migration. However, whether it means waiting until there is more data available, or making better use of what already is, the enormous global significance of the topic re quires that much more effort be allocated towards exposing the particulars of the ICT migration relationship
55 Appendix : Table 1 Dummy Variable Frequencies Dummy Variable N % = 1 Std. Dev. Migrate (Dependent Variable ) 1246280 5 5882 0.229694 Occupation type 1246280 33 2849 0.499373 Age 30 39 1246280 26 7799 0.442813 Age 40 49 1246280 25 6393 0.436641 Age 50 59 1246280 19 1115 0.39318 Age 60 69 1246280 8 6244 0.280724 Age 70 89 1246280 2 1612 0.145414 High School Diploma 1246280 16 5838 0.371936 1 3 Years of College and no Bachelor's Degree 1246280 23 2986 0.422734 4 Year of College or Bachelor's Degree 1246280 16 2558 0.368962 5+ Years of College or Graduate Degree 1246280 10 5768 0.307541 Table 2 Macroeconomic Control Means Macroeconomic Control N Mean Std. Dev. Unemployment Rate 1246280 6.624282 1.778945 Labor Force Participation Rate 1246280 64.83692 2.077672 Percent Change in GDP 1246280 2.565666 2.149501
56 Table 3 Analysis of Maximum Likelihood Estimates (All Years) Parameter Estimate St. Error Wald Chi Sq Pr>ChiSq Occupation type 0.062 0.000244 64693.3 <.0001 Age 30 39 0.9167 0.000249 13575800 <.0001 Age 40 49 1.4651 0.000297 24257005 <.0001 Age 50 59 1.8347 0.000376 23752356 < .0001 Age 60 69 2.1199 0.00061 12091853 <.0001 Age 70 89 2.5441 0.00151 2846118 <.0001 High School Diploma 0.0555 0.000347 25612.46 <.0001 1 3 Years of College and no Bachelor's Degree 0.01437 0.000302 226320.6 <.0001 4 Year of College or Bachelor's Degree 0.3833 0.000341 1266886 <.0001 5+ Years of College or Graduate Degree 0.4421 0.000394 1261494 <.0001 Labor Force Participation Rate 0.00608 0.000053 13247.89 <.0001 Unemployment Rate 0.0461 0.000063 541397.7 <.0001 Percent Change in GDP 0.0255 0.000053 232912.7 <.0001 Table 4 Odds ratio Estimates (All Years) Effect Point Estimate 95% Wald Confidence Limits Occupation type 1.064 1.063 1.064 Age 30 39 0.4 0.4 0.4 Age 40 49 0.231 0.231 0.231 Age 50 59 0.16 0.16 0.16 Age 60 69 0.12 0.12 0.12 Age 70 89 0.079 0.078 0.079 High School Diploma 0.946 0.945 0.947 1 3 Years of College and no Bachelor's Degree 1.155 1.154 1.155 4 Year of College or Bachelor's Degree 1.467 1.466 1.468 5+ Years of College or Graduate Degree 1.556 1.555 1.557 Labor Force Participation Rate 0.994 0.994 0.994 Unemployment Rate 0.955 0.955 0.955 Percent Change in GDP 1.026 1.026 1.026
57 Table 5 Mean Value of Dependent Variable for Non Macroeconomic Independent Variables Independent Variable Mean of Dependent Variable (Migration Rate Mean) 1968 1968 1978 1979 1989 1990 2000 2001 2012 2012 Occupation Type ICT 7.92203 7.54103 7.47319 6.71949 4.12131 3.765 Non ICT 5.17078 5.20655 6.09976 5.61986 3.89409 3.511 Age 14 29 16.8835 15.7425 14.5212 14.4152 10.7834 9.824 30 39 7.44246 6.78405 6.77385 6.50616 4.46491 4.229 40 49 3.90522 3.67908 4.32063 4.21052 2.55946 2.643 50 59 2.1753 2.25071 2.94371 2.98969 2.05873 1.849 60 69 1.65965 1.89068 2.27586 2.06631 1.64056 1.280 70 89 1.02041 1.13944 0.97738 1.46816 1.21843 1.474 Education No High S chool Diploma 4.7 4.678 5.31 4.705 3.54625 3.2 High S chool Diploma 7.0959 8.30822 7.82438 5.61067 3.63541 3.352 1 3 Years of College and no Bachelor's Degree 8.7381 8.0644 7.69182 6.67740 4.06931 3.591 4 Years of College or Bachelor's Degree 10.585 10.6748 9.52455 8.36187 4.60732 4.046 5+ Years of College or Bachelor's Degree 12.0466 10.0830 8.25297 6.87162 4.15163 4.142 Table 6 Mean Value of Dependent Variable for Macroeconomic Independent Variables Macroeconomic Controls Mean of DV (all years) Unemployment Rate < 6% 5.7683638 Unemployment Rate 6% or > 6% 5.665533 Labor Force Participation Rate < 65.5% 5.5495414 Labor F orce Participation Rate > 65.5% 6.288575 Change in GDP < 3% 5.00105 Change in GDP > 3% 6.203365
58 Table 7 Linear Regression Parameter Estimates for dummy IV's YEAR Intercept Age (Compared to Age 14 29) 30 39 40 49 50 59 60 69 70 89 1968 0.15568 0.095 0.12825 0.14312 0.14933 0.15518 1969 0.15071 0.09549 0.12178 0.13794 0.13342 0.14689 1970 0.14708 0.09316 0.12162 0.13441 0.1404 0.14731 1971 0.14765 0.09344 0.12376 0.13379 0.14049 0.14626 1976 0.14005 0.08491 0.11264 0.12296 0.13043 0.13951 1981 0.12594 0.07405 0.0969 0.11043 0.11581 0.12916 1982 0.13644 0.08243 0.10921 0.12185 0.12429 0.13426 1983 0.11963 0.06455 0.09156 0.10545 0.11083 0.11976 1984 0.13782 0.08135 0.10818 0.12375 0.12292 0.13405 1986 0.14046 0.09124 0.10857 0.12268 0.1267 0.1477 1987 0.13489 0.07801 0.10389 0.11142 0.11604 0.13071 1988 0.14169 0.08442 0.10947 0.12019 0.12869 0.13676 1989 0.13291 0.082 0.10093 0.10993 0.12509 0.1333 1990 0.14995 0.08906 0.11657 0.13353 0.14512 0.15368 1991 0.14641 0.09234 0.11114 0.12417 0.13522 0.14724 1992 0.12317 0.07184 0.09711 0.1064 0.11472 0.11265 1993 0.13875 0.08494 0.10841 0.11859 0.1263 0.12712 1994 0.13757 0.07799 0.106 0.11493 0.12618 0.13517 1996 0.14494 0.08618 0.11848 0.12346 0.1318 0.14095 1998 0.12949 0.08597 0.10771 0.1165 0.122 0.12743 2000 0.15521 0.0923 0.12019 0.13522 0.14141 0.15192 2002 0.14492 0.0843 0.1152 0.1258 0.13119 0.13988 2004 0.13098 0.07952 0.10746 0.11587 0.11842 0.1299 2006 0.1229 0.06536 0.09305 0.10086 0.11015 0.11312 2008 0.09837 0.06287 0.08526 0.08827 0.09776 0.08339 2009 0.10458 0.06004 0.08072 0.0908 0.0915 0.09856 2010 0.09099 0.05736 0.0777 0.08254 0.08291 0.08713 2011 0.09719 0.06897 0.08276 0.09008 0.0957 0.09845 2012 0.10745 0.0614 0.082 0.08926 0.09858 0.09539
59 Table 8 Linear Regression Parameter Estimates for dummy IV's (Continued) YEAR Education (Compared to No HS Diploma) Occupation (Compared to non ICT) HS Diploma 1 3 Years of College 4 Years of College or Bachelor's Degree 5+ Years of College or Graduate Degree ICT 1968 0.00515 0.02347 0.04192 0.05978 0.00971 1969 0.02019 0.01331 0.04446 0.04932 0.00701 1970 0.00803 0.01932 0.05679 0.04187 0.0063 1971 0.02004 0.01006 0.04053 0.03067 0.00719 1976 0.00435 0.00928 0.03603 0.02429 0.01003 1981 0.00181 0.0166 0.03314 0.03137 0.00416 1982 0.01512 0.01306 0.04018 0.03311 0.00124 1983 0.00732 0.00914 0.029 0.02972 0.00501 1984 0.02903 0.01145 0.03295 0.03299 0.00195 1986 0.01051 0.00871 0.03338 0.03088 0.01328 1987 0.00186 0.01691 0.04756 0.03594 0.00171 1988 0.01064 0.009 0.03457 0.02214 0.0064 1989 0.00643 0.01841 0.03522 0.02604 0.00843 1990 0.0086 0.0097 0.03277 0.03271 0.00792 1991 0.01247 0.0114 0.03943 0.02507 0.00033868 1992 0.00456 0.01876 0.03349 0.03248 0.00109 1993 0.00103 0.01158 0.03219 0.02727 0.00235 1994 0.00564 0.0031 0.02758 0.01576 0.00564 1996 0.0081 0.00739 0.02503 0.0224 0.00387 1998 0.0021 0.01713 0.02723 0.0354 0.0049 2000 0.00109 0.00788 0.02471 0.01987 0.0021 2002 0.00002063 0.00587 0.02062 0.01942 0.00061705 2004 0.00722 0.01459 0.02715 0.02735 0.00339 2006 0.00336 0.00217 0.00897 0.012 0.0005263 2008 0.00443 0.00923 0.01583 0.01582 0.00287 2009 0.00283 0.00387 0.00917 0.01321 0.00289 2010 0.00687 0.01125 0.01607 0.01646 0.00286 2011 0.00783 0.00903 0.01559 0.01801 0.00239 2012 0.000462 0.00383 0.0807 0.01362 0.00010937
60 Table 9 Logistic Regression Odds Ratio Estimates for Age (Compared to Age = 14 29) Year 30 39 40 49 50 59 60 69 70 89 1968 0.39 0.21 0.124 0.093 0.066 1969 0.371 0.219 0.122 0.148 0.075 1970 0.375 0.209 0.13 0.095 0.061 1971 0.37 0.191 0.127 0.089 0.06 1976 0.403 0.228 0.161 0.113 0.063 1981 0.421 0.263 0.168 0.125 0.036 1982 0.399 0.229 0.148 0.125 0.065 1983 0.461 0.226 0.165 0.121 0.062 1984 0.41 0.242 0.143 0.144 0.073 1986 0.371 0.268 0.182 0.157 0.031 1987 0.425 0.262 0.207 0.176 0.078 1988 0.407 0.255 0.187 0.134 0.089 1989 0.404 0.285 0.224 0.127 0.075 1990 0.407 0.254 0.157 0.095 0.045 1991 0.37 0.26 0.181 0.113 0.044 1992 0.438 0.268 0.201 0.139 0.156 1993 0.388 0.246 0.181 0.131 0.126 1994 0.418 0.241 0.182 0.111 0.056 1996 0.392 0.199 0.166 0.117 0.0063 1998 0.367 0.231 0.177 0.137 0.101 2000 0.395 0.238 0.158 0.122 0.065 2002 0.404 0.215 0.155 0.12 0.068 2004 0.402 0.221 0.171 0.151 0.076 2006 0.442 0.234 0.179 0.111 0.088 2008 0.391 0.198 0.166 0.097 0.205 2009 0.405 0.221 0.135 0.131 0.066 2010 0.398 0.204 0.16 0.157 0.113 2011 0.337 0.218 0.157 0.113 0.088 2012 0.417 0.243 0.183 0.11 0.134
61 Table 10 Logistic Regression Odds Ratio Estimates for Education (Compared to Education = No High School Diploma) Year 5 or more years of college or a graduate degree 4 Years of College or Bachelor s Degree 1 3 Years of College HS Diploma 1968 2.232 1.779 1.451 0.982 1969 2.004 1.848 1.263 1.359 1970 1.92 2.176 1.395 1.191 1971 1.646 1.782 1.21 1.353 1976 1.454 1.6 1.181 1.111 1981 1.684 1.653 1.326 1.01 1982 1.736 1.795 1.26 1.291 1983 1.669 1.585 1.195 0.91 1984 1.715 1.634 1.225 1.531 1986 1.616 1.571 1.164 1.196 1987 1.81 1.951 1.33 1.064 1988 1.451 1.613 1.17 1.198 1989 1.551 1.661 1.354 1.139 1990 1.681 1.576 1.184 1.166 1991 1.554 1.783 1.223 1.24 1992 1.904 1.824 1.465 1.141 1993 1.688 1.688 1.252 0.996 1994 1.373 1.535 1.171 0.906 1996 1.55 1.5 1.158 0.851 1998 2.034 1.688 1.439 1.075 2000 1.429 1.461 1.149 1.02 2002 1.458 1.416 1.114 0.992 2004 1.741 1.656 1.33 1.147 2006 1.315 1.199 1.046 0.923 2008 1.555 1.497 1.286 1.123 2009 1.449 1.269 1.107 0.913 2010 1.656 1.59 1.398 1.229 2011 1.738 1.574 1.337 1.285 2012 1.435 1.223 1.106 0.986
62 Table 11 Logistic Regression Odds Ratio Estimates for Occupation Type (Compared to Occupation = non ICT) Year ICT intensive/ White Collar Occupation Type 1968 1.188 1969 1.131 1970 1.122 1971 1.173 1976 1.182 1981 1.074 1982 1.02 1983 1.09 1984 1.03 1986 1.231 1987 0.97 1988 1.107 1989 1.143 1990 1.126 1991 1.002 1992 1.014 1993 1.034 1994 1.097 1996 1.062 1998 1.09 2000 1.033 2002 1.007 2004 0.935 2006 1.007 2008 1.077 2009 0.921 2010 0.922 2011 1.067 2012 0.999
63 Bibliography The Changing American Family. 2001. The New York Times May 18, 2001, sec Opinion. Altshuler, Rosanne, and Williams, Robertson. 5 myths about your taxes. in Tax Policy Center [database online]. 2010 [cited 2/19/2013 2013]. Available from http://www.taxpolicycenter.org/publications/url.cfm?ID=901335 Bartel Ann P. 1979. The migration decision: What role does job mobility play? The American Economic Review 69 (5) (December): 775 -786. Clark, W. A. V. 1986. Human M igration 1st ed. Beverly Hills: SAGE Publications, Inc. Comin, D. A., and B. Hobijin. 2004. Cross country technological adoption: Making the theories face the facts Journal of Monetary Economics : 39 83. Cooke, Thomas. 201 2 Internal migration in decline. The Professional Geographer Doi: 10.1080/00330124.2012.7243 43 It is not just t he economy: Declining migration and the rise of secular rootedness. Population, Space and Place 17 (7 March 2011): 193 203. Cooke, Thomas J. 2003. Family Mi gration and the relative earnings of husbands and wives. 93 (2): 338 49. DaVanzo, Julie. 1978. Doe s unemployment affect migration? Evidence from micro data. The Review of Economics and Statistics 60 (4) (Nov.): 504 14. 1978. Does unemployment affect migration? Evidence from micro data. The Review of Economics and Statistics 60 (4) (Nov.): 504 14. Ferreira, Fernando, Joseph Gyourko, and Joseph Tracy. 2010. Housing busts and household mobility. Journal of Urban Economics 68 (1): 34 45. Fisc her, Claude S. 2002. Ever more R ooted Americans. City & Community : 177. FRED, Federal Reserve Economic Data, Federal Reserve Bank of St. Louis. 2012. Civilian labor force participation rate (CIVPART) U.S. Department of Labor: Bureau of Labor Statistic 2012. Civilian unemployment rate (UNRATE) U.S. Department of Lab or: Bureau of Labor Statistics
64 2 012. Gross domestic product (GDPA) U.S. Department of L abor: Bureau of Labor Statistic Freund, Rudolf J. and Ramon C. Littell. 2000. SAS System for Regression 3rd ed. Cary, NC: SAS Institute Inc. Green, A. E., T. Hogarth, and R. E. Shackleton. 1999. Lo ng distance commuting as a substitute for migration. International Journal for Population Geography 5 (1): 49 67. Greenwood, Michael J., Gary L. Hunt, Dan S. Rickman, and George I. Treyz. 1991. Migration, regional equilibrium, and the estimation of compen sating differentials. The American Economic Review 81 (5) (Dec.): 1382 90. Grigg, D. B. 1977. E. G. R avenstein and the "laws of migration" Journal of Historical Geography 3 (1): 41 54. Gross, Emily. Internal revenue service area to area migration data: S trengths, limitations, and current trends. Internal Revenue Service : 185 90. Hardill, Irene. 2002. Gender, migration, and the dual career household. Hardill, Irene, and Anne Green. 2003. Remote working -altering the spa t ial contours of work and home in t he new economy. New Technology, Work and Employment 18 (3): 212 22. Ioannides, Yannis M., Henry G. Overman, Esteban Rossi Hansberg, and Kurt Schmidheiny. 2007. The effect of information and communication technologies on urban structure Panel Meeting of Ec onomic Policy in Lisbon : 1 37. Jasper, James. 2000. Restless nation: Starting over in A merica Chicago, IL: University of Chicago Press. Julsrud, Tom Erik, Randi Hjorthol, and Jon Martin Denstadli. 2012. Business meetings: Do new videoconferencing techno logies change communication patters? Journal of Transport Geography 24 (September 2012): 396 403. Kan, Kamhon. 1999. Expected and unexpected residential mobility. Journal of Urban Economics 45 (1) (1): 72 96. Khwaja, Yasmeen. 2000. Should I stay or should I go? M igration under uncertainty: A new approach ed. University of London. School of Oriental and African Studies. Dept. of EconomicsUniversity of London. King, Miriam, Steven Ruggles, J. Trent Alexander, Sarah Flood, Katie Genad ek, Matthew B. Schroeder, Brandon Trampe, and Rebecca Vick. 2010. Integrated public
65 use microdata series, current population survey: Version 3.0. [machine readable database] Minneapolis, MN: Minnesota Population Center [producer and distributer]. Lee, E. S. 1969. A theory of migration Migration 56 : 282 97. Long, Larry. 1988. Migration and residential mobility in the United States New York: Russell Sage Foundation. Molloy, Raven, Christopher L. Smith, and Abigail K. Wozniak. 2011. Internal migration in the united states. National Bureau of Economic Research (17307). Morrison, Donna Ruane, and Daniel T. Lichter. 1988. Family migration and female employment: The problem of underemployment among migrant married women. Journal of Marriage and Family 50 (1 ): 161 72. Morrison, Philip S., W. A. V. Clark, Kirsten Nissen, and Robert Didham. 2010. Moving for employment reasons California Center for Population Research 24 : 1 20. Mueser, Peter, and Philip E. Graves. 1995. Examining the role of economic opportunity and amenities in explaining population redistribution. Journal of Urban Economics 37 (1): 1 25. Partridge, Mark D., Dan S. Rickman, M. Rose Olfert, and Kamar Ali. 2012. Dwindling U.S. internal migration: E vidence of spatial equilibrium or structural shifts in local labor markets? Regional Science and Urban Economics Popnoe, David. 1985. Private pleasure, public plight; American metropolitan community life in comparative perspective Vol. 1st ed. New Bruns wick, NJ: Transaction Books. Roof, Wade Clark, and William McKinney. 1987. American mainline religion New Brunswick: Rutgers University Press. Sasser, Alicia. 2010. Voting with their feet? local economic condit ions and migration patterns in N ew England. New England Public Policy Center Shauman, Kimberlee A. 2009. Migration and changing family characteristics in the U.S. 1981 2005. Department of Sociology, University of California Davis Sjaastad, Larry A. 1962. The costs and returns of human migration Journal of Political Economy 70 (5, Part 2: Investment in Human Beings) (Oct.): 80 93. Trost, Katherine. Video conferencing adoption: Tracking trends and deployment strategies. in Search Unified Communication [database online]. 2010 [cited January 24 20 13 2013]. Available from
66 http://searchunifiedcommunications.techtarget.com/feature/Video conferencin g adoption Tracking trends and deployment strategies U.S. Census Bureau. Population estimates terms and definitions. in U.S. Census Bureau [database online]. U.S. Census Bureau, 2012 [cited October 29 2012]. Available from http://www.census.gov/popest/about/terms.html 2011. Movers by type of move and reasons for moving: 2009. U.S. Census Bureau, 2007. Renters four times more likely to move than homeowners. 2001. Why people move: Exploring the march 2000 current population survey. U.S. Census Bureau, P23 204. Wolf, Douglas A. 2005. Our "increasingly mobile society"? The curious persistence of a false belief. The Gerontologist 45 (1): 5 11. Wuthrow, Robert. 1994. Sharing the journey New York: Free Press. Zabel, Jeffrey. 2009. The role of the housing market in migration response to employment shocks. New England Public Policy Center (09 2).