Reducing automation risk through career mobility: where and for whom?

Automation risk prevails less in large cities compared to small but little is known about the drivers of this emerging urban phenomenon. A major challenge that automation quantified by work-related tasks allows for measurement through occupation, which turn implicitly related local economic structure and individual career paths. This paper examines role working on changes mobility. Using panel data Swedish workers, we show metropolitan effect reducing mainly induced inter-firm job Separate estimates different groups accrues mostly native, high-skilled male workers. El riesgo de automatización prevalece menos en las grandes ciudades que pequeñas, pero se sabe poco sobre los factores impulsan este fenómeno urbano emergente. Una principales dificultades es el cuantifica mediante tareas relacionadas con trabajo permiten la medición a través profesión, su vez está relacionada implícitamente estructura económica y trayectorias profesionales individuales. Este artículo explora papel del cambios movilidad Se utilizaron datos trabajadores suecos, mostró efecto metropolitano reducción inducido principalmente por laboral entre empresas. Las estimaciones separadas para distintos grupos muestran corresponde todo nativos, altamente cualificados sexo masculino. オートメーション(自動化)のリスクは小都市に比べて大都市では少ないが、この新たな都市現象の要因についてはほとんど解明されていない。オートメーションのリスクは職業による測定を可能にする仕事に関連するタスクによって定量化されるのであるが、逆にそれは、地域の経済構造や個人のキャリアパスに絶対的に関連するということが大きな課題である。本稿では、個人の雇用流動性によるオートメーションのリスクの変化における都市部で働くことの役割を検討する。スウェーデンの労働者に関するパネルデータを用いて、オートメーションのリスクを低減する大都市効果は、主に企業間の雇用流動性によって惹起されることを示す。異なる集団の各々の推定から、この効果の恩恵の大部分は、地元出身でスキルが高い男性の労働者が受けていることが示される。 influences labour markets replacing human workforce certain (Autor et al., 2003; Brynjolfsson & McAfee, 2011). Whether threatens existing jobs or facilitates creation new depends type investments into automation, worker skills their potential renewal (Acemoglu Restrepo, 2020a). Because technologies as well skill level typically differ across regions, problem has necessary spatial dimension. Accordingly, workers have been found face higher risks than (Crowley 2021; Frank 2018). Further evidence suggests regions lead employment growth long run because robots supplement automatable trigger upgrades (Leigh 2020). Nevertheless, how individuals adopt avoid threat geography process one important questions quickly evolving, still largely uncovered, field (Frank 2019). In paper, expand recent aggregate analyses (e.g., Crowley 2021) detailing micro-mechanisms explaining agglomeration relation regional vulnerability. Specifically, look at prevents risks. We argue create more favourable conditions mobility, are able reduce exposure automation. The therefore lies facilitating upward According central tenets economics, two mechanisms take place regard. First, demand non-automatable high due functional specialization (Duranton Puga, 2005). Second, arenas learning offer additional opportunities advance perform difficult replace (De Roca 2017; Glaeser Maré, 2001; Gordon, 2015). However, not homogenous ability learn cities, arguably base other personal attributes may determine High-skilled example, can further upgrade while low-skilled enjoy (MacKinnon, 2017). empirical analysis exploits employer-employee matched covering 10% random sample over 2005–2013 period. apply occupation-level measure introduced Frey Osborne (2017) quantify much an exposed Descriptive statistics help us understand getting low-risk job. Further, multinomial logit models occupation mobility enables examine helping careers prevent technology-driven displacement. Our results there stark differences between concerning distribution high- jobs. During period, Sweden experienced remarkable rise share concentrated with growing intensity. largest accounted 58% overall 2005 2013, was 65% case Workers significantly chances job, holds when controlling unobserved heterogeneity. find decrease differently, depending sex level. Evidence implies (at least college degree) realize dynamic suggesting networking helps them risk. female low educated (less immigrants these effects relatively smaller. finding it easier skilled automation-proof metro areas provide some support Moretti's (2012) idea factors behind selection ongoing clustering ‘innovative’ After introductory section, proceeds four steps. next section provides review literature followed explanation our estimation strategy Section 3. 4 introduces sets stage individual-level econometric 5. then ends concluding remarks. As advances fields change everyday life, prospective market have, unsurprisingly, become subject public concern. OECD report, Nedelkoska Quintini (2018) 46% (ca 210 million) all 32 countries 50% chance being replaced machines near future. At same time, divergence increasing characterize development many developed economies (Iammarino 2019; Storper, accentuates challenges sustaining welfare non-core regions. Although broad discussions society premise “end-of-work” stressing adverse matter academic debate will indeed influence markets. While widespread adoption artificial intelligence, robotics computer-aided manufacturing imply displacement performing “rule-based” routine 2018, Autor, 2014; Autor 2011; Heyman, 2016; Levy Murnane, 2004), others emphasize increases high-skill complementary programming, maintenance, design), seemingly unrelated non-tradable activities culture, arts, leisure, services) inter-industry spillovers Salomons, 2018; Moretti, 2010). disruptive technological innovations future work independently from location however rarely addressed question. Some studies using robot recently showed production-related 2020a, 2020b; Graetz Michaels, absence detailed limits understanding automation-driven unfold economy, findings suggest terms motivation such might be present. Research antecedents locational patterns job-creation could vary space hence affect differently location. Since occupations manage control remotely codified rules, they tend located outskirts rural areas, where land rents wages usually lower (Moretti, 2012; Scott, 2009; 2013). both creative low-skill service catering childcare geriatric nurses, janitors, cleaners, hairdressers etc.) (Scott, 2009). Additionally, tends confined larger activities, innovation capabilities knowledge competences already present (Muneepeerakul 2013; Shutters 2016). Consequently, general displacement, likely hit harder via tasks, remain resilient even benefit positive on-going (Andersson 2020; Eriksson Hane-Weijman, study US statistical corroborates perspective showing average measures frequently adopted calculated (see e.g., Arntz Osborne, 2017), al. similar Henning (2016) (2021) European NUTS regions) should interpreted caution. “automation risk” geographical unit (or industry) compresses information occupational mix hides within itself. Given region primarily shaped reflect set location-specific also associated industry structure. Information communication (ICT) firms rationalize production space, move offshore keep only managerial supporting services cities. do require educational attainment social considered “bottlenecks” 2004). less-automatable who expected better starting refer static moving city gets irrespective whether search precedes vice versa. Static shifts cause one-off reduction allowing relative ease. 11 Gordon (2015) labelled impact elevator effects. Therefore, one-time cannot pure “city-effects” stem simply demand. Another way contributes prevention step ahead shorter period time. Career progression particularly access highly-valued elements tacit professional networks (Gordon, That is, cutting-edge skills, gathering experience top ranked employers making valuable business connections long-lasting career-path prospects Hence, gain wider range positions, entrepreneurs (Faggio Silva, 2014). vertical along ladder develop time accumulation capital Recently, De Puga provided indirect faster wage growth. 22 D'Costa Overman (2014) premium young performed levels providing expertise, managing resources, negotiating, interfacing customers, interpersonal types (Borghans 2008), accompanies Note underlying within-city diverse downward fact concentrate Rather, are, part, consequence Glaeser, 1999; 2015), effective information-exchange (Ioannides Loury, increased early-career turnover (Wheeler, 2008). grasped upgrades. properly extent risk, rule out determining structure, especially Controlling heterogeneity extremely probability abilities. shown Heckman (2006) others, non-cognitive abilities personality traits preferences path Taking characteristics account examining backgrounds degrees. sense necessarily protected technology-induced unemployment extent. argued Sicherman Galor (1990), among adapt various situations. (college degree more) primary beneficiaries Besides capital, however, structural workers' family background, household division labour, discrimination, These behaviour, substantial women men, native By identify susceptible begin simple conceptual framework linked choices. task-content jobs, most movements More precisely, comes involves occupation. any factor options assumed result intertemporal utility maximization seek sequence form optimal path. Expected lifetime function earnings choice constrained traits, education, external worker. For intra-firm uncertain within-firm transfers promotions employers’ decision, whereas availability opportunities. point choose options. Then, determines direction temporal worker’s convenient career-upgrading periods did type, removing “move-years” sample. doing so eliminate migration focus happen during stay type. 55 follows premium. mobile suitable climb (van Ham 2001), representative whole workforce. presence fixed resulting s referred population. Nonetheless, estimating Equation (3) excluding observations years best 77 Estimating drawback partial evaluated without specifying odds-ratio Of course, approach used category or, indeed, full city-region dummies. figures Figures 1 2) three few specifications consider multiple categories (metros city-regions). use Statistics 2013. data, annual active market, occupations, residence municipality recorded (along like age, sex, status, etc.). chosen reasons (c.f., Eriksson, 2021): previous accounts identified polarization tendencies prior initial year analysis, makes purpose expect occur. final revision codes 2013 onwards comparisons after virtually impossible. To define 4-digit SSYK-96 nomenclature broadly consistent international ISCO-88 classification. allocated municipalities address workplace. classified region-types; “Metro areas” labels (Stockholm, Gothenburg Malmö), “Large areas’ urbanized university Umeå, Linköping, Karlstad, etc. “Small regions” refers remaining, sparsely populated, Sweden. Originally, distinguishes five first second correspond “Metro” “Large” categories, rest up “Small” category. number predefined fixed-effects ML identification relies observing outcomes (as discussed 3). considering dummies would reasonable amount data. Table A1 Appendix, almost half population resides (49% 2013) 52% employees. It greatest taken (11.8%), employees declined 1% income smallest accompanied increase incomes. Manufacturing industries closely connected machine operators assemblers belonging SSYK-codes 7 8) withdrawn pace region-types requiring degree. (Henning 2021). signals often administrative units (employing assemblers) greater smaller looked Department Labour’s O*NET database assess computerization 702 6-digit Standard Occupational Classification (SOC). 88 proposed mixed methodology expert evaluations methods. step, group AI experts were asked hand-label 70 assigning if automatable, 0 not. authors several variables describe perception manipulation, creativity, intelligence required occupations. train algorithm probabilistic assessment SOC (including test occupations). Estimates available appendix (2017). scores mapped SOC-10 onto classification official correspondence tables SOC-00, SSYK-96. neither conversion one-to-one correspondence, correspondences averaged out. result, 352 (out 355) given score. dataset offers task content its limitation substantial. Most significantly, actual possible another staying considerable individual. Moreover, acquire firm 99 Especially, firm-specific assets. , 1010 sensitive threshold, ran model values (i.e 25 percentage points) led conclusion. does involve coefficient workplace ( ) capture spectrum 2, those horizontal lists bottom distribution. less-susceptible include medical doctors, psychologists, special education teaching professionals, therapists nurses. Clearly, caring, perceptiveness. physical skills. contains clerks. considerably what kind 2 short list common low- high-risk each region-type. upper panel, sales (such computer designers, programmers high-education professionals) preschool primary-level activities. Conversely, forms assemblers, concerns clerks (lower panel). depicted Figure 1, disparities concern concentration employment, Comparing indications remaining fairly stable ones. increasingly partly destruction Sweden, attributed last decades (Eriksson (right) peripheral north-western border parts (left). Summary reported 3 movers (i.e., once region-types) stayers region-type). models, region-type based (for 3), see systematic stayers. Only 12% observed changed mover younger stayer (43 years). 16 34 thus geographically (Lundholm, 2007). “35–44 years” age sample, “25–34 years.” 48% 43% child. Along slight seen Among fewer parents. One-third