The outbreak of the COVID-19 pandemic this year has led to a surge in teleworking and caused renewed interest in studying the importance of commuting patterns for geographical labour markets. Geographical labour markets are usually denoted as commuting zones, and many economic phenomena are studied at the commuting zone level. An important aspect of a commuting zones are commuting patterns, which are determined by travel costs and time for workers commuting from remote areas. However, advanced transportation technology and information communication technology (ICT) may decrease the importance of such commuting costs, and the ‘death of distance’ or the ‘persistence of distance’ has been discussed in the literature for years. To examine this, a useful definition of the geographical unit of a labour market is imperative.
Geography as a unit of analysis
Geography is used as a unit of analysis in a wide range of empirical analyses. In order to elucidate the underlying economic mechanisms, many studies rely on differences across geographic units such as regions. One example is Autor et al. (2013), who study the impact of the China shock on the US. In order to understand how increased competition due to products imported from China affected US employment, they compare regions with a high share of employment in industries which faced increasing competition from China – such as textiles, light industrial products – to other regions. In such an analysis, the appropriate choice of geographical unit as the unit of analysis is crucial.
Administrative units such as municipalities or prefectures are often selected as geographical units of analysis in many studies of the Japanese economy. This is convenient as a lot of data is usually assembled on administrative unit levels. However, administrative units may not always capture local economic activity. In a case where many workers commute across municipalities for example, a wider geographical unit than the municipality might be appropriate.1 However, if a bigger geographic unit – such as prefectures, for example – is chosen as the unit of analysis, there is a concern that effects differ across labour markets within the same prefecture which may lead to large errors in the estimates. A further consideration when choosing the appropriate unit of analysis are potential spillover effects. When considering local employment policies, for example, if companies are attracted by tax incentives to a certain municipality causing many workers to commute from neighbouring municipalities, these spillover effects to neighbouring municipalities should be taken into account when analysing such a policy. For this reason, a broader geographical unit may be appropriate when considering employment policies.
Methodology for creating commuting zones
To deal with these problems of regional units, we create regional units using commuting zones (CZ). The key idea is that if there are many workers who commute between different municipalities, we assign those cities to the same commuting zone. We hence create a set of labour markets within which workers are likely to commute. In the US, many analyses have already been conducted using commuting zones as a geographical unit,2 usually relying on the same methodology for creating CZs. In our study we rely on micro data from Japan's Population Census3 conducted by the Ministry of Internal Affairs and Communications to construct commuting zones. Compared to the US, Japanese people tend to live much closer together as the Japanese population is about half of the US’ population but it is spread on an area of only 1/26 of that of the US. Further, commuting across administrative units is much more common in Japan than in the US. For these reasons, defining a regional unit is trickier in the Japanese case compared to the US case.
Figure 1 shows the map of our defined commuting zones for Japan for the year 2015. We constructed 265 commuting zones from1,736 municipalities. We defined an area as a commuting zone if the percentage of workers commuting within that are is high. On average, the percentage of commuting within the same geographic unit is 62% for municipalities, 87% for our defined commuting zones, and 91% for prefectures. Further, Figure 2 shows the distribution of geographical units by calculating the ratio of commuting within the same unit for each of the 1,736 municipalities, 265 commuting zones, and 47 prefectures. The horizontal axis in the figure measures the percentage of commuting happening within the same unit. If we focus on geographical units where 90% of commuting happens within the same unit, we see that this is the case in only less than 10% of all municipalities. In contrast to this, it is the case in 90% of prefectures and 80% of commuting zones. Our measure of a commuting zone is hence quite good as it has a similar rate of within-unit commuting as prefectures, but our defined commuting zones are much smaller. It can further be seen that municipalities do not seem to be an appropriate geographical unit to capture labour markets. Over 50% of all municipalities experience less than 70% of all commuting within the municipality.
Figure 1 Japanese commuting zones in 2015
Figure 2 Proportion of workers who commute within a geographic unit
(a) Population unweighted
b) Population weighted
Validity of our commuting zones to capture labour markets
Finally, we show the results of a variance analysis which demonstrates the effectiveness of our commuting zones in capturing heterogeneity on the labour market. We decompose the variances of characteristic variables of labour markets such as the employment rate or wages into the variance within an area and the variance between areas after controlling for personal attributes. Specifically, by using microdata from the Basic Survey of Wage Structure, we measure how the residual dispersion becomes smaller after controlling for several types of regional dummy variables. We confirm that the largest part of the variance of the employment rate and the wage are indeed explained by the dummy variable of the commuting zones. In other words, labour markets are heterogeneous across different commuting zones but homogeneous within the same commuting zone. We are hence confident that our commuting zones effectively capture labour markets.
As a commuting zone is a combination of municipalities, and administrative data is collected by municipalities, it is straightforward to create commuting zone-level data. In order to make our commuting zones available for future academic use, we share the concordance of commuting zones from 1980 to 2015.4
Authors’ note: The correspondence of standardized municipality code and the CZ code is available at https://github.com/daisukeadachi/commuting_zone_japan. For further details see Adachi et al. (2020).
Acemoglu, D and P Restrepo (2020), "Robots and jobs: Evidence from US labour markets", Journal of Political Economy 128(6): 2188-2244.
Adachi, D, T Fukai, D Kawaguchi and Y U Saito (2020), “Commuting Zones in Japan,” RIETI Discussion Paper Series 20-E-021.
Autor, D, D Dorn and G H Hanson (2013), "The China syndrome: Local labour market effects of import competition in the United States", American Economic Review 103 (6): 2121-68.
Chetty, R, N Hendren, P Kline and E Saez (2014), "Where is the land of Opportunity? The Geography of Intergenerational Mobility in the United States", The Quarterly Journal of Economics 129 (4): 1553-1623.
Tolbert, C and M Sizer (1996), "US commuting zones and labour market areas: A1990 update," Technical Report.
Kanemoto, Y and K Tokuoka (2002), “Nihon no toshiken settei kijun (Proposal for the Standards of Metropolitan Areas of Japan)”, Ouyou chiikigaku kenkyuu (Journal of Applied Regional Science) 7: 1-15.
1 According to the 2005 Population Census, only around 60% of commuting activity happens within the same municipality.
2 The method to create commuting zones by Tolbert and Sizer (1996) is based on the clustering method of “Hierarchical agglomerative clustering (HAC)” which is a basic method of machine learning, where a similarity measure is defined depending on the degree of commuting between municipalities and a set of municipalities with high similarity are then combined recursively.
3 A similar concept as that of a commuting zone (CZ) in Japan is the urban employment area (UEA) defined by Kanemoto and Tokuoka (2002). UEAs are similar to CZs as they are created depending on commuting patterns between municipalities. However, UEAs are created to capture the commuting areas from which workers commute to core cities. Hence, many more remote areas do not belong to any UEA. Contrary to this, CZs created by the above mentioned clustering method are mutually exclusive and cover all municipalities.
4 Note that many municipalities are merged from 1980 to 2005 and the geographical unit of a municipality is changing. We create versions of commuting zones to deal with this problem.