The effects of transport
infrastructure investments
on commuting
MASTER
THESIS WITHIN: Economics NUMBER OF CREDITS: 30
PROGRAMME OF STUDY: Economic analysis AUTHOR: Dennis Johansson
JÖNKÖPING May 2021
Master Thesis in Economics
Title: The effects of transport infrastructure investments on commuting - in Sweden Authors: Dennis Johansson
Tutor: Almas Heshmati Date: 2020-06-16
Key terms: Transport, Infrastructure, Investments, Labor, Commuting, Microeconomics
Abstract
The purpose of this paper was to investigate the effects of transport infrastructure investments on commuting. This is an exciting topic because when the transport infrastructure becomes more developed in an economy, it can also increase the accessibility for labor, enabling them to take advantage of more labor opportunities and even lower unemployment. A fixed-effects model was estimated, and the variables included were chosen after revising literature and data availability. The data used in the estimation of the model comes from Statistics Sweden, it covers all the 290 municipalities in Sweden, and the data is for the years 2011-2019. The results from the regression models found a significant effect for transport infrastructure investments on commuting, and the effect was found to be instant rather than lagged. These findings suggest that there is a relationship between investments into the transport infrastructure and the number of individuals that commute both into and out of municipalities. The results could mean that the investments have enabled labor to allocate themselves more efficiently when it comes to their workplace and that individuals can access more labor opportunities. The implication from the concluding results is that Swedish municipalities most likely have policymakers that properly handle investments into the transport infrastructure as labor may reap the benefits of the investments in the same period they are made.
Table of Contents
1. Introduction ... 1
2. Literature Review ... 4
2.1 Importance of transport infrastructure ... 4
2.2 Accessibility ... 5
2.3 Commuting ... 6
2.4 Lagged effect for investments ... 7
2.5 Summary ... 8
3. Methodology ... 9
3.1 Hypothesis ... 9
3.2 Data & Variables ... 9
3.3 Descriptive statistics & correlation matrix ... 11
3.4 Model specification & estimation ... 13
4. Empirical results ... 16
5. Discussion ... 20
6. Limitations ... 22
7. Conclusion ... 23
8. References ... 25
9. Appendix ... 28
Tables
Table 1. Abbreviations, definitions & expected effects of variables. ... 11
Table 2. Descriptive statistics of all variables. ... 11
Table 3. Correlation table. ... 13
Table 4. Results from regression models ... 17
Table 5. Breusch and Pagan Lagrangian multiplier test... 28
Table 6. Hausman test ... 28
1. Introduction
The infrastructure of an economy is a crucial component needed for the economy to function in the long run, where the infrastructure can be everything from the roads that one drives on to the tap water that one has in their home. The part of the infrastructure that this paper will focus on is the transport infrastructure where roads, railroads and parking are included. Many governments have for a long-time used investment into the infrastructure to increase the country's economic growth, and there has been a vast amount of research done surrounding this subject, where some research has been done in the matter of transport infrastructure.
The research question that is to be answered in this paper is, have investments into the transport infrastructure of Swedish municipalities increased the amount of labor that commute to and work in municipalities? This is an interesting topic because when the transport infrastructure does become more developed in an economy, it also increases the accessibility for labor so that they can come and work within areas that have more labor opportunities if their current place of residence, for example, cannot provide enough jobs. This may allow labor within an economy to allocate themselves more efficiently than if the transport infrastructure were not developed enough, and it can also lower the unemployment rate.
The purpose of this paper is to learn more about how transport infrastructure investments might affect labor allocation for Swedish municipalities. The study in this paper will be a quantitative one where the data used is in panel form, so it studies different municipalities over time. A fixed-effects model is used in the estimation, where included variables have been chosen by reviewing literature and after data availability. The concluding results can be used by policymakers in the future when looking into how to allocate government funds for investments in municipalities.
There has been a lot of research done surrounding the role of transport infrastructure investments and especially how public transport has a vital role in society. Glaeser et al., (2008) findings suggest that individuals who might not be able to transport themselves without public transport will choose to live in connection with public transport access. But they also found that individuals with higher incomes will decide to allocate themselves outside of cities because they can afford private transport. Blind et al., (2017)
found that new commuting opportunities for individuals born in a non-western country but who lived in Sweden increased their earnings and employment probability. This is consistent with the findings of Glaeser et al., (2008) that individuals who might be “poorer” are affected more by transport infrastructure than what individuals with higher earnings might be. Nevertheless, individuals who still live outside of the areas where they work and use private transport are still affected by changes in the transport infrastructure as functioning roads is an essential factor.
Living costs are also significant for individuals when they choose where to live and work, as the gains from living in a specific area might not always exceed the costs of living in the area. A solution to this problem is for labor to live in a different place than they work in, and so they can choose to commute to work which was suggested by findings from Cameron and Muellbauer (1998). Their findings also stated that commuting costs go up with time and distance which are factors that can be changed through investments into the transport infrastructure so labor can afford longer trips if the infrastructure improvements lower the time factor and so total cost.
Sweden is a developed country where most of the population does already live-in urban areas compared to rural areas as a study from Statistics Sweden (n.d.b) showed that in 2018, 87% of the population lived in urban areas, but that does not mean that every urban municipality is as big as the other or have as many labor opportunities. Larger municipalities do still attract more labor than smaller municipalities, which can be connected to agglomeration economies where economic activities are clustered in certain areas, which can be seen in many Swedish cities. But the cost of living inside cities in Sweden can also be high, so the transport infrastructure is essential as many individuals choose to commute to their job. Sweden is also a country that has historically had a steady amount of immigration where it started going up in the year 2006 and in 2016, it was at its highest level ever, even though since then it has gone down to a lower level (SCB, n.d.a). Immigrants are a group of individuals who may rely on public transport when transporting themselves, so infrastructure investments are most likely necessary for that group of society, which was also found by Blind et al. (2017).
The data that will be used to conduct the research in this paper comes from SCB (Statistics Sweden, 2020), the data collected is for all 290 municipalities in Sweden and the data covers the years from 2011 to 2019.
Results from the estimated regression models do show that transport infrastructure investments do have a significant and positive effect on commuting. These results do answer the research question in this paper, and it does confirm it as well. Furthermore, the effect is found to be instant rather than lagged, which shows that the population can reap the benefits of the investments in the same period as they happen. As the effect is instant, policies towards transport infrastructure investments have most likely been well planned. Thus, policymakers could evaluate why investment policies during the time-period have been functional so that they might be able to replicate the results in the future. The rest of the paper is organized as follows: in section 2, a literature review will be presented that contains the relevant theory and literature for the research being conducted, section 3 contains the methodology of the paper, section 4 contains the empirical results, section 5 includes a discussion surrounding the results, in section 6 some limitations are presented, and finally in section 7 a conclusion is written.
2. Literature Review
In this part of the paper literature that is relevant for the research being conducted will be both presented and reviewed. The literature included will be surrounding the subjects of transport infrastructure, labor, accessibility, and commuting.
2.1 Importance of transport infrastructure
Glaeser et al., (2008) investigated the phenomenon why lower-income individuals choose to locate themselves in cities, and it was found that a reasoning behind it was that they choose to reside in connection to public transport systems. Meanwhile, other findings also suggested that higher-income individuals decide to locate themselves so that they can use private transport, so they do not necessarily have to live and work in the same area. Thus, the findings by Glaeser et al., (2008) indicate that transport infrastructure is vital for both high- and low-income individuals.
Accessibility to commuter trains for labor is investigated by Blind et al. (2017), where it is stated that the economic position of labor can be stronger through access to public transportation as it can give them wider access to jobs which in turn may allow the labor market to function better. It is also stated that if commuting is possible for labor, then reservation wages may be lowered, which can lower unemployment. A reservation wage is a minimum wage that a worker would accept to participate in the labor market. Further findings by Blind et al. (2017) suggest that individuals born in a non-western country but currently lived in Sweden had increased earnings through the introduction of new commuting opportunities, and the chance of being employed increased. However, this group of individuals does, on average, earn less than normal income and might not have access to other means of transportation.
In the paper by Meersman and Nazemzadeh (2017), the authors state that developed transport infrastructure can increase the economic growth of an economy, even though it is also stated that the role of transport infrastructure investments plays to improve the economic growth is not always clear, as it is a complex mechanism. The researchers found no evidence that transport infrastructure investments do directly affect GDP, but some evidence was found that the length of motorways and the port infrastructure could have effects on GDP in the long run.
Some researchers bring forward the important point that transport infrastructure needs to be well-planned in economies, and policies towards investments need to be well thought out. Costa and Markellos (1997) state that it is crucial to investigate the efficiency of transport infrastructure as only looking at the output is not enough due to it being a public service. Looking at efficiency is helpful as it may allow policymakers to see if the transport infrastructure is being well handled. Crafts (2009) also talks about the importance of having good policies towards public capital investments as bad policies may lead to congestion and longer travel times for road traffic. Then Hull and Karou (2014) states that the growth of urban areas has made it harder for specific individuals to make their daily commutes as they might not have access to private transport. This shows that it is important to have well-planned transport networks, especially public transportation networks, that are easily accessible for the population.
Fingleton and Szumilo (2019) investigated how investments into the transport infrastructure may affect wages. Their findings showed that wage levels are correlated with travel times, indicating that connections to other locations can be important for local wages. The results could then suggest that transport infrastructure investments can increase salaries even though the increases might be spatially uneven. The gains were highest at locations where commuting times went down the most. This is because regions may get access to labor that before the transport infrastructure investments were not able to commute.
2.2 Accessibility
In the paper by Maroto and Zofio (2016), the authors summarize and describe how economic geography and regional economics are related to accessibility and transport infrastructure. In economic geography, accessibility is related to the mobility of both labor and firms, which may lead to agglomeration and dispersion forces, in the case of agglomeration, it means that economic activities are centralized, while dispersion forces mean that economic activities are more spread out. Both agglomeration and dispersion forces are then, in turn, are affected by transportation costs. In regional economics, increases in accessibility may lead to economic development in the long run, and it may reduce income inequality in regions. Maroto and Zofio (2016) state that improvements of the transport infrastructure can change the access to different markets, which may change
the distribution of economic activities. Some findings by the authors suggest that transport infrastructure investments may increase economic accessibility.
Further connections between accessibility and transport infrastructure investments are found by Rokicki and Stepniak (2018), which suggest that increases in accessibility through investments into the transport infrastructure can be related to regional employment growth. But it was also negatively related to output growth in rural areas, showing that if labor have access to more labor opportunities, then that can have a negative effect in rural areas because individuals might change their workplace with increased accessibility.
The findings by Matas et al. (2015) state that investments into the transport infrastructure can have a positive effect on productivity, and that can be seen as an increase in agglomeration. Some further findings by the authors showed that wages may be increased through increases in accessibility.
Rouwendal (1998) findings suggest that unemployment may be lower in areas where individuals can access job opportunities in more than one employment center compared to if only one employment center would be available. So, more accessibility to jobs for individuals can lead to lower unemployment.
Meijer and Rouwendal (2001) connects accessibility with both commuting and housing, where it is found that increased accessibility for labor is crucial as it allows workers to travel the same distance in a shorter time or a longer distance for the same amount of time. This is important, as according to Meijer and Rouwendal (2001), labor has a strong preference for housing, and if the accessibility of labor is good, then that allows individuals to choose their place of residence in a better manner due to commuting access.
2.3 Commuting
Roberts and Taylor (2017) summarize how labor economics looks at commuting. It is described in such a way that individuals will choose their place of residence in a manner where one looks at price, the location, and amenities. Commuting comes into the equation as it can be used by labor to travel to their workplace, then longer commutes can be compensated by higher wages or better working conditions. To expand on the subject, one can look at the paper by Haas and Osland (2014), where the willingness to commute is investigated, and their findings suggest that it may depend on the wage structure in
combination with affordable and accessible housing. Haas and Osland (2014) also state that mobility is a dynamic process shaped by both labor and housing markets.
Monte et al., (2018) also investigates more on how commuting relates to the housing market. The authors state that if commuting opportunities are available for labor, then that may allow workers to separate their workplace and place of living. Further findings by Monte et al., (2018) suggest that if there would be a demand shock for labor in an area, then the increases in employment would be more significant if commuting were more available for labor compared to if it were not. It is stated by the authors also that commuting is vital for the spatial distribution of economic activities.
In the paper by Dodson and Li (2020), there are some interesting findings on commuting costs, where it was found that the growth of employment centers was correlated with transport costs. These findings show that labor might be willing to trade higher commuting costs for increases in employment opportunities. Further findings by Dodson and Li (2020) suggest that it is essential for governments to use infrastructure investments to improve both housing, work connections and transport costs.
2.4 Lagged effect for investments
Several researchers have investigated different type of investments into the infrastructure and have found the same result that there might be a lagged response for the investments. Berechman et al., (2006) investigated whether there was a connection between transport infrastructure investments and economic development. The findings by Berechman et al., (2006) suggest that investments into public capital have a significant positive effect on output both at a state and on a county level. Their research indicates that the output might react with a lag to some public capital investments while some other public capital investments might respond instantly.
There are more researchers with the same lagged result, such as Brage-Ardao et al., (2013) who found that investments into the transport infrastructure do have a stronger effect in the long run compared to the short run. The largest effect was found to be for investments into the road infrastructure. The findings by Brage-Ardao et al., (2013) suggests that the benefits from investments may take several years to be reflected in the output.
Another researcher who investigated transport infrastructure, or more specifically, rail transportation networks, is Cats (2017). They state that there is a lag when transportation
networks are being developed. It is a long process, as it takes time from the planning, to then follow up on investment decisions, where the investments are used for construction and finally, the launch of the transportation network. Thus, according to Cats (2017), it may take time for the population to reap the benefits of the investment as it may take a long time for the results from the investment to show up.
Then there are researchers such as Folmer et al., (2019) who investigated the accessibility gains in the economy from the opening of a tunnel which is an investment into the infrastructure. It was found that the effect from the investment came instantly as the tunnel opened, and even some changes happened before the opening, meaning that there might have been an anticipation of the opening from the population. Even though Folmer et al., (2019) does not necessarily mention how long the construction of the tunnel might have taken after the initial investment.
2.5 Summary
What can be learned from the reviewed literature is that the objective for commuting is both a minimization and maximization problem simultaneously. The minimization of commuting can be for less energy spent, total cost, and time usage. While the maximization of commuting can be for productivity increases, increased employment opportunities and lower unemployment. Thus, the objective for commuting is a combination of both, where transport infrastructure investments can assist with both minimization and maximization. In minimization, transport infrastructure investments may allow individuals to travel with shorter times, and it may lower the cost of travelling. Then in the case of maximization, transport infrastructure investments may let productivity go up as it can enable individuals to allocate themselves more efficiently, leading to increases in employment opportunities. If individuals can take advantage of more employment opportunities, then unemployment can go down.
What has been learned in the literature review will be used in the following section to choose which variables that will be included in the regression model.
3. Methodology
In this part of the paper, the hypothesis, the variables used in the regression model, the data with descriptive statistics and the regression model will be presented.
3.1 Hypothesis
According to theory and previously done research, investments into the transport infrastructure should increase the mobility of labor and allow them to take advantage of more employment opportunities and so the workers in the economy might be able to allocate themselves more efficiently. Therefore, the main hypothesis in this paper is have investments into the transport infrastructure increased the number workers that commute both to and out of municipalities.
3.2 Data & Variables
The gathered data in this thesis comes from Statistics Sweden (SCB) and is in panel form, so it is multi-dimensional, meaning that it studies different subject over time. Hence, it is a combination of time series with cross-sectional data. The starting year for the data is 2011, and the end year is 2019, so the data spans for a time of 9 years. The data collected is for all the 290 municipalities in Sweden, so there will be 290 observations for each year in the study, and in total, it will be 2610 observations.
Total Commuting (TC)
The dependent variable Total Commuting is the amount of labor that both commute into a municipality and the amount of labor that commute out of a municipality. Total commuting is used as labor do both travel out of and into municipalities, so the total amount needs to be included to see the total effect from the independent variables. What is included in the data for each municipality is the number of individuals that live in the municipality and commute to their job in another area, and then also the number of individuals that commute into the municipality to their job from other areas. An important note as well is that every individual does show up two times in the data as both an in-commuter and out-in-commuter. The variable cannot measure the separate effects of either out-commuting or in-commuting that would be measurable if only one of the parts would have been used.
Transport infrastructure investment (TII)
The independent variable Transport infrastructure investment is the number of funds that a given municipality has invested into transport infrastructure. This is the independent variable that the focus will be on in this paper as it will be answering the research question and the hypothesis. The variable is expected to positively affect the dependent variable because it is expected that when municipalities invest more funds into the transport infrastructure, that can allow more individuals to commute to that municipality and work there. In addition, it may also enable individuals to commute out of the municipality to other areas. The value of the variable is in thousands of Swedish crowns.
Average Earnings (AE)
The independent variable Average Earnings is the average amount of earnings that individuals have in the municipality. This is the first control variable that will be used, and it is used in the regression model because when wages go up in an area, then that area will be more attractive for labor, so individuals might choose to commute to these places for work rather than moving their place of residence there. Therefore, the variable is expected to have a positive effect on total commuting. The value of the variable is in thousands of Swedish crowns.
Average House Price (AHP)
The independent variable Average House Price is the average cost to buy a house in the given municipality. This is the second control variable that will be used in the regression model. This variable is included because when house prices are high in an area, then individuals might choose to have their workplace there but not their place of residence in the same area, so they can choose to commute instead. Therefore, the variable is expected to have a positive effect on total commuting. The value of the variable is in thousands of Swedish crowns.
Labor Supply (LS)
The independent variable Labor Supply is the number of employed and over 16 years old who reside in a municipality but do not necessarily have their workplace there. This is the third control variable used in the regression model. It is included because when the total number of people who live and are employed in a municipality increases, then total
commuting can increase. Therefore, the variable is expected to have a positive effect on total commuting.
Below in table 1, a list of the key variables is provided along with their definitions and the independent variables expected effect on total commuting.
Table 1. Abbreviations, definitions & expected effects of variables.
Abbreviation Definitions Expected effect
TC Total Commuting in municipality i at time t
TII Transport infrastructure investment in municipality i at time t
+
AI Average Income in municipality i at time t + AHP Average House Price in municipality i at time t +
LS Labor Supply in municipality i at time t +
3.3 Descriptive statistics & correlation matrix
To get an overview of how the dataset looks like descriptive statistics will be provided for all the variables where the main variables total commuting and transport infrastructure investments will be discussed, and other variables might also be discussed if there are any interesting values that stand out. The descriptive statistics can be found below in table 2. Table 2. Descriptive statistics of all variables.
Variables Mean Median Std. dev. Min Max
TC 10981 4590 29289 306 460782
TII 65110 24427 175901 485 2243681
AE 262 255 38 189 568
LS 16437 7464 35857 1054 522215
Even though the data in the descriptive statistics is just combined for all the years, which is because that including descriptive statistics for all the years separately would take up too much space, and it would not necessarily be of more value for the paper.
Looking at the variable TC, one can see that the standard deviation is larger than the mean, implying that some values are large outliers compared to the smaller values in the data. It is also clear that even though it is spread out over several years, there are some municipalities that experience very low commuting when looking at the minimum value, and there are some municipalities that experience a large amount of commuting when looking at the maximum value.
Then when looking at the TII variable, one can see that the case for that variable is the same as for TC as the mean value is lower than that of the standard deviation, which implies that there are some big outliers in the data. By looking at the minimum value, one can see that there are some municipalities that do not invest a lot into the transport infrastructure, and by looking at the maximum value, one can see that some municipalities invest a great amount of funds into the transport infrastructure. Even though one municipality could invest a large amount one year and then in the next year, it could be investing less.
One outtake from the descriptive statistics is that four of the variables have considerable standard deviations, which could be harmful when estimating the model later, as one would want the variables to be normally distributed. One solution to this is that the variables are used in their logarithm form in the regression model instead of their normal values. This may allow for the variables to be more normally distributed, as significant outliers can make the distribution of the variables skewed.
Table 3. Correlation table. TC TII AE LS AHP TC 1 TII 0,8652*** 1 AE 0,2905*** 0,1702*** 1 LS 0,9483*** 0,9440*** 0.2247*** 1 AHP 0,4835*** 0,3543*** 0,8343*** 0.4070*** 1 ***Significant at the 1% level
By looking at the correlation table one can see that there are some variables that are highly correlated with one another, every variable is significant at the one percent level meaning that they are correlated with one another.
Most notably, the transport infrastructure investment variable, which is the main independent variable, is highly correlated with the dependent variable, which is interesting as one would want the main independent variable to be correlated with the dependent variable. One can also see that there might be two possible multicollinearity problems that may cause issues later when the regression model is going to be estimated, as the variable transport infrastructure investments is highly correlated with the labor supply variable and the average earnings variable is highly correlated with the average house price variable. But this could also be because of how the data is gathered for the municipalities, as it could be that certain municipalities have the same sort of outliers in the data, which then shows up as a correlation between the variables. This issue will be discussed further in the empirical results. The correlation between the variables would look different if the variables were in their log form instead of the standard form, which the variables will most likely be when estimating the regression model later.
3.4 Model specification & estimation
To be able to answer the hypothesis with the use of data a regression model will be estimated, Gujarati & Porter (2009) present some different models that can be used for panel data where the one that most likely will fit the data in this paper is that of a fixed effects model (FEM).
A more general version of the fixed effects model would look like:
𝑌𝑖𝑡 = 𝛽1𝑖+ 𝛽2𝑋𝑖𝑡+ ⋯ + 𝛽𝑘𝑋𝑖𝑡+ 𝑢𝑖𝑡 (1)
The dependent variable Y could, for example, be entity i at time t, so i is the cross-sectional aspect in the model while t is the time series part of the model and k stands for the number of independent variables. There are different intercepts for all the entities and the slopes for the variables are the same for all entities while the coefficients are also constant over time.
By rewriting the fixed effects model with the variables in this paper then the regression model would look like this:
ln 𝑇𝐶𝑖𝑡 = 𝛽1𝑖+ 𝛽2ln 𝑇𝐼𝐼𝑖𝑡+ 𝛽3ln 𝐴𝐸𝑖𝑡+ 𝛽4ln 𝐴𝐻𝑃𝑖𝑡 + 𝛽5ln 𝐿𝑆𝑖𝑡+ 𝑢𝑖𝑡 (2)
Where TC is total commuting and the dependent variable, 𝛽1𝑖 is the intercept for each of the municipalities, TII is transport infrastructure investments and the main independent variable, AE is the average earnings variable, AHP is the average house price variable, LS is the labor supply variable and 𝑢 is the error term. The variables will also be in their log form during the estimation as the distribution of the variables were found to be highly skewed.
Now, of course, one cannot assume that the FEM model will automatically be the best one to use, so to know which model is best, two tests need to be performed to choose a model. Firstly, a Breusch and Pagan Lagrangian multiplier test will be performed to select between a standard pooled OLS model and a random-effects model (REM). The result will most likely favor the REM model in favor of the pooled OLS. Then a Hausman test will be performed to see if a FEM is in favor of the REM model. The test will also most likely favor the FEM model as it allows for the different municipalities in this paper to have different intercepts that most likely are not random.
When estimating the model, what will also be necessary is that every problem that can be found in both cross-sectional models and in time series models can be found in panel data models. Therefore, several tests must be done to check for multicollinearity, heteroskedasticity and autocorrelation. After these tests have been done, a proper solution must be applied to the regression model if the tests show any shortcomings in the model. Multicollinearity is when one or more of the explanatory variables are correlated with one
another, heteroskedasticity is present when the error term in the regression model is non-constant, and autocorrelation is present when the errors or disturbances in a model is not random or correlated.
From the literature reviewed before, some researchers (Berechman et. al., 2006; Brage-Ardao et. al., 2013; Cats, 2017) mentioned that there might be a lag in the economy for the effects from investments into the transport infrastructure to show up. Because of this, a regression model will be estimated that does not include a lag for the variable transport infrastructure investments, and another regression will be estimated that does have a lag for the variable transport infrastructure investments. By doing this, the results from the regressions will indicate if the effect from the variable is instant or if the effect is lagged.
4. Empirical results
In this part of the thesis, the results from the regression model will be presented. Three models have been estimated where one of the resulting models includes the labor supply variable while the other two does not. This is because there might be a multicollinearity problem present when the labor supply variable is included. Before one could choose to use a fixed effect model the proper tests had to be done to see if it were the best fit for the research being conducted. So, a Breusch pagan Lagrangian multiplier test was performed to see if a pooled OLS model would be preferred over a random-effects model and the p-value for the chi test was 0,0000 (See table 5 in the appendix), which means that the null hypothesis was rejected in the test, meaning that a random-effects model is preferred over a pooled OLS model. Then a Hausman test was performed to see if the fixed effects model was to be preferred over a random-effects model. In this test, the p-value was 0,0000 (See table 6 in the appendix), meaning that once again, the null hypothesis was rejected, so the fixed-effects model is preferred over a random-effects model.
Then after the proper panel data model was chosen the results from the models could be estimated with the statistical software. Tests were performed for both heteroskedasticity and autocorrelation in the model. The null hypothesis was rejected in both cases, which means that both heteroskedasticity and autocorrelation are present in the models. To solve this problem the standard errors in the models has been corrected as estimates are not biased if the model is estimated correctly its only inefficient which is why robust standard errors are being used as it solves both the problem of autocorrelation and heteroskedasticity. This is a solution that was proposed by both Greene (2012) and Verkbeek (2008).
The result from the empirical model is presented below in table 4, the statistical software Stata was used to estimate the fixed effects models.
Table 4. Results from regression models Variables Including LS (1) Including LS with robust SE (1) Excluding LS (2) Excluding LS with robust SE (2) Excluding LS with robust
SE & TII lag (3) Log (TII) 0.0002 (0.0046) 0.0002 (0.0038) 0.0092* (0.0048) 0.0092** (0.0047) 0.0055 (0.0057) Log (AE) 0.3383*** (0.0274) 0.3383*** (0.0314) 0.4908*** (0.0275) 0.4908*** (0.0368) 0.5084*** (0.0462) Log (AHP) 0.0227** (0.0106) 0.0227* (0.0119) 0.0445*** (0.0113) 0.0445*** (0.0124) 0.0414*** (0.0135) Log (LS) 0.6527*** (0.0383) 0.6527*** (0.0961) Constant 0.5446* (0.2964) 0.5446 (0.7798) 5.3805*** (0.0920) 5.3805*** (0.1512) 5.3433*** (0.2007) R2 0.5699 0.5699 0.5159 0.5159 0.4584
Standard errors in parentheses. ***Significant at the 1% level **Significant at the 5% level *Significant at the 10% level
Three different regression models have been estimated where all are using robust standard errors, and the results from two of the models without robust standard errors are also provided to show the difference compared to when robust standard errors are used. The first regression model is the one that contains the variable labor supply, in the second model, the labor supply variable has been excluded as there most likely is a multicollinearity problem present in the first model and the third model uses a lag for the variable transport infrastructure investments.
Firstly, one can look at the goodness of fit for all the models, or in more straightforward terms, how well do the independent variables explain the variations in the dependent variable. This can be seen in the r squared value where the first model does have a value of 0.5699, the second model does have a value of 0.5159, and the third model does have
a value of 0.4584. This implies that all the models can roughly explain around 50 percent of the variations in the dependent variable total commuting. If one were to pick a model based on this value, then the first model would seem to be the best one, but as there might be a multicollinearity problem present, then the first models’ results might not be reliable. Robust standard errors have been applied to all the models because, as mentioned before, autocorrelation and heteroskedasticity was present in all regression models. But there is no such solution to the multicollinearity problem. One can see that the value, for example, of the constant changes a lot in the regression results when the labor supply variable is included even though the value of the constant will not be the main point in the discussion as every municipality has its own intercept in a fixed-effects model. Furthermore, the value of the coefficient for transport infrastructure investments also changes quite a lot from the first to the second model, which implies further that there might be a multicollinearity problem. As a fixed effects model does not suffer from omitted variable bias, it is then better to omit the variable labor supply, the results from the first model will not be presented further as it is seen to be unreliable, and the second and third model can thus be seen as more trustworthy.
There was also the chance of a second multicollinearity problem, as the variables average earnings and average house price where highly correlated. To then check if one of the variables had to be excluded, the regression was estimated three times where one only has the variable TII, the second one has excluded the AHP variable and the third one excludes the AE variable. The results can be found in table 7 in the appendix and the value for the full model without the LS variable is included as well for easy comparison. There are some differences that can be found between the coefficients for the estimated regressions, where the most interesting differences are between the models that either exclude AHP or AE. The one that excludes AHP is very close to the full model which includes both variables, compared to the model which excludes AE that has changes in the level of significance for the TII variable and the value of the coefficients for both TII and AHP almost doubles. If one then must choose between either including both variables or excluding one of them, then the inclusion of both would seem to be a correct choice. Because if one starts with the model that excludes AHP and then adds the variable back, then there are not any drastic changes in the values for the coefficients, standard errors, significance levels or the R2.
The primary independent variable in this paper is transport infrastructure investments. The variable is found to be significant at the 5 percent significance level in the second model. The variable is not significant at any significance level in the third model. The regression was estimated with the variables in their log form, making the fixed effects model a log-log model. This makes it easier to interpret the values of the coefficients. Thus, for the second model, for a one percent increase in the transport infrastructure investments, the total commuting is expected to go up by 0,0092 percent if everything else stays constant.
Furthermore, the variables average earnings and average house prices are significant at the one percent significance level in both the second and the third model. In the second model a one percent increase in the average earnings variable is expected to increase total commuting by 0,4908 percent and a one percent increase in the average house price variable is expected to increase total commuting by 0,0445 percent. The same goes for the third model where a one percent increase in average earnings is expected to increase total commuting by 0,5084 percent and a one percent increase in average house price is expected to increase total commuting by 0,0414 percent.
5. Discussion
Firstly, one can discuss the main difference between the second and third models. The independent variable transport infrastructure investments were only significant in the second model and not the third one. This implies that there is no evidence of lagged effect for investments into the transport infrastructure. Instead, there is an instant effect present, which is interesting as some researchers stated that there might be a lagged effect from the investments into the transport infrastructure (Berechman et. al., 2006; Brage-Ardao et. al., 2013; Cats, 2017). Even though Berechman et al., (2006) and Folmer et al., (2019) also stated that the effect from investments into the transport infrastructure might be instant. Thus, the effect of transport infrastructure investments in the economy is not too surprising. It may be that Swedish municipalities are rather open with the planning of the transport infrastructure, so the workers who use it to transport themselves might be able to anticipate changes when they happen.
Now, if it is the case that individuals can anticipate when the changes in the transport infrastructure will happen, then that would most likely mean that the municipalities have well-planned transport infrastructure. This follows what Crafts (2009) stated in his paper that there need to be good policies towards the transport infrastructure for it to function in the long run.
The evidence of the effect is still a bit surprising as, for example, Blind et al. (2017) did not find any significant effects for the population when looking into new commuting opportunities for individuals unless they went and investigated a minority group. Also, as Meersman and Nazemzadeh (2017) stated that the role of transport infrastructure is a rather complex mechanism in the economy which is why there might have been hard to find evidence of the effect before.
Nevertheless, the hypothesis in this paper can then be confirmed in the sense that investments into the transport infrastructure have been found to have a significant effect on commuting, which means that it has increased the number of individuals that do commute both out of municipalities and into municipalities. It is important to note that the model does look for changes over time. Hence, the effects show that transport infrastructure investments have positively affected total commuting over time.
This may also imply that transport infrastructure investments have increased labor mobility, so individuals' accessibility might have gone up, making it possible for them to take advantage of more labor opportunities.
The control variables for average earnings and average house prices were also found to be having a positive and significant relationship with commuting. These effects were expected as reviewed literature stated that individuals choose their place of residence due to several factors such as the price of the housing, the location, and the available amenities (Roberts & Taylor, 2017). Average earnings are connected to the location as areas with a higher amount of earnings can be seen as more attractive to labor, and so they might experience more in-commuters. This is because individuals might try to be employed in locations where they can receive a higher salary even though they live in a different area. Then house prices relate to commuting in the sense that some areas might be expensive to live in, so the benefits do not outweigh the costs, but still, individuals might choose to work there, so they live in another area and commute to work (Monte et. al., 2018). The results then in this paper for the control variables does confirm what was stated by the authors. So, it strengthens the literature surrounding commuting in connection with earnings and housing.
6. Limitations
This paper has some limitations, and the first one is that when looking into transport infrastructure investments, one would want to have data covering an extended time. So even though this paper covers data for nine years, it could be interesting if that number would have been higher.
Investments into the transport infrastructure in this paper look strictly at a municipal level, but many public capital investments are not necessarily only done on a municipal level. It could also be on, for example, a county level.
There might be a reverse causality issue for the variables used in the fixed-effects model. That is, the dependent variable could be causing the independent variables. Therefore, there is a possibility that the variable total commuting could be affecting transport infrastructure investments.
The variable labor supply had to be excluded due to a multicollinearity problem. However, one could have used a ratio of total commuting divided by labor supply, making it so that labor supply would have ended up in the intercept. Even though that would also change the primary dependent variable and the other coefficients, but it could still be an exciting topic for future research.
7. Conclusion
In this paper, the effects of transport infrastructure investments on commuting were to be investigated. The empirical results from the regression model do not reject the hypothesis that investments into the transport infrastructure have a positive relationship with total commuting. Therefore, the research question has been answered in this paper, meaning that investments into the transport infrastructure is positively correlated with total commuting.
This result was unexpected in the sense that previously, it has been hard for researchers to find an effect from transport infrastructure investments. The relationship between the variables was found to be significant on the five percent significance level. Thus, this paper contributes to the literature about transport infrastructure investments and labor. The increases in commuting may mean that individuals can mobilize themselves more efficiently than if the investments into the transport infrastructure were not present. The results also give an idea that Swedish municipalities are now using good policies towards investments into the transport infrastructure as the effect is instant rather than lagged, meaning that individuals can reap the benefits of investments instantly, which was distinguished by using a lag in one model and no lag in another one. It would most likely then be in the best interest of the municipalities to continue with the same policies that have been used during the nine years if their goal is to increase commuting by improving the transport infrastructure. If for example policymakers have other goals with investments and do not want to increase commuting, then they might need to oversee their work and change how funds are being invested. As there might be cases where you would not want to increase in-commuting as it may, for example, cause congestions, which then again is why policymakers need to oversee their work constantly to see that investments continue to be efficient and follow the correct goals. What seems to be important as well from reviewed literature is that policymakers need to still think of the bigger picture, so they should create a well-working transport infrastructure as it is essential for the economy to function well in the long run.
There are some topics that could be interesting for future research. For starters one could try to find data for immigrants in Sweden to see if the relationship between commuting and transport infrastructure investments are even stronger for that part of the society, as
some literature states that immigrants are affected more by transport infrastructure investments. One could also differentiate between different types of transport infrastructure investments, for example it would be interesting to either investigate only commuting by public transport or one could see if private transport has been limited by the amount of parking spots that exist in employment centres. Performing research on a county level rather than a municipal level could also be interesting as all investments into the transport infrastructure are not strictly done on a municipal level.
It could also be interesting to use a more sophisticated econometric model in future research, to see if the results would be different compared to the fixed effects model used in this paper, as the fixed effects model used is not the most complex model.
8. References
Berechman, J., Ozbay, K., & Ozmen, D. (2006). Empirical analysis of transportation investment and economic development at state, county and municipality levels.
Transportation, 33(6), 537-551.
Blind, I., Dahlberg, M., & Åslund, O. (2017). All aboard? Commuter train access and labor market outcomes. Regional Science and Urban Economics, 67, 90-107.
Brage-Ardao, R., Graham, D. J., & Melo, P. C. (2013). The productivity of transport infrastructure investment: A meta-analysis of empirical evidence. Regional Science and
Urban Economics, 43(5), 695-706.
Cameron, G., & Muellbauer, J. (1998). The housing market and regional commuting and migration choices. Scottish Journal of Political Economy, 45(4), 420-446.
Cats, O. (2017). Topological evolution of a metropolitan rail transport network: The case of Stockholm. Journal of Transport Geography, 62, 172-183.
Costa, Á., & Markellos, R. N. (1997). Evaluating public transport efficiency with neural network models. Transportation Research Part C: Emerging Technologies, 5(5), 301-312.
Dodson, J., & Li, T. (2020). Job growth, accessibility, and changing commuting burden of employment centres in Melbourne. Journal of Transport Geography, 88, 102867. Fingleton, B., & Szumilo, N. (2019). Simulating the impact of transport infrastructure investment on wages: A dynamic spatial panel model approach. Regional Science and
Urban Economics, 75, 148-164.
Folmer, K., Hoogendoorn, S., Van Gemeren, J., & Verstraten, P. (2019). House prices and accessibility: evidence from a quasi-experiment in transport infrastructure. Journal
of Economic Geography, 19(1), 57-87.
Glaeser, E. L., Kahn, M. E., & Rappaport, J. (2008). Why do the poor live in cities? The role of public transportation. Journal of Urban Economics, 63(1), 1-24.
Greene, W. (2012). Econometric Analysis. 7th Edition, Prentice Hall, Upper Saddle River.
Gujarati, D. N., & Porter, D. (2009). Basic Econometrics. Mc Graw-Hill International Edition.
Haas, A., & Osland, L. (2014). Commuting, Migration, Housing and Labour Markets: Complex Interactions. Urban Studies, 51(3), 463–476.
Hull, A., & Karou, S. (2014). Accessibility modelling: predicting the impact of planned transport infrastructure on accessibility patterns in Edinburgh, UK. Journal of Transport
Geography, 35, 1-11.
Maroto, A., & Zofío, J. L. (2016). Accessibility gains and road transport infrastructure in Spain: A productivity approach based on the Malmquist index. Journal of Transport
Geography, 52, 143-152.
Matas, A., Raymond, J. L., & Roig, J. L. (2015). Wages and accessibility: The impact of transport infrastructure. Regional Studies, 49(7), 1236-1254.
Meersman, H., & Nazemzadeh, M. (2017). The contribution of transport infrastructure to economic activity: The case of Belgium. Case Studies on Transport Policy, 5(2), 316-324.
Meijer, E., & Rouwendal, J. (2001). Preferences for housing, jobs, and commuting: a mixed logit analysis. Journal of Regional Science, 41(3), 475-505.
Monte, F., Redding, S. J., & Rossi-Hansberg, E. (2018). Commuting, migration, and local employment elasticities. American Economic Review, 108(12), 3855-90.
Roberts, J., & Taylor, K. (2017). Intra-household commuting choices and local labour markets. Oxford Economic Papers, 69(3), 734-757.
Rokicki, B., & Stępniak, M. (2018). Major transport infrastructure investment and regional economic development–An accessibility-based approach. Journal of Transport
Geography, 72, 36-49.
Rouwendal, J. (1998). Search theory, spatial labor markets, and commuting. Journal of
Urban Economics, 43(1), 1-22.
Statistics Sweden. (n.d.a). Invandring till Sverige. https://www.scb.se/hitta-statistik/sverige-i-siffror/manniskorna-i-sverige/invandring-till-sverige/
Statistics Sweden. (n.d.b). Tätorter i Sverige. https://www.scb.se/hitta-statistik/sverige-i-siffror/miljo/tatorter-i-sverige/
9. Appendix
Table 5. Breusch and Pagan Lagrangian multiplier test Test: Var(u) = 0
chibar2(01) = 8339.77 Prob > chibar2 = 0.0000
Table 6. Hausman test
Test: Ho: difference in coefficients not systematic chi2(3) = 920.41
Prob>chi2 = 0.0000
Table 7. AE & AHP, Multicollinearity regressions.
Variables Exclude AE & AHP
Exclude AHP Exclude AE Full model
Log (TII) 0.0685*** (0.0125) 0.0096** (0.0047) 0.0177*** (0.0055) 0.0092** (0.0047) Log (AE) 0.5873*** (0.0233) 0.4908*** (0.0368) Log (AHP) 0.2248*** (0.0095) 0.0445*** (0.0124) Constant 7.826344*** (0.1279) 5.16394*** (0.1334) 6.7101*** (0.0842) 5.3805*** (0.1512) R2 0.0454 0.5127 0.4494 0.5159
Standard errors in parentheses. ***Significant at the 1% level **Significant at the 5% level *Significant at the 10% level
