The impact of metropolitan technology on the non-metropolitan labour market: evidence from US patents

ABSTRACT

While urban technology exerts a positive effect on rural development through knowledge spillovers, it also raises the competitive advantage of urban firms over rural firms in product market competition. Urban technology also affects the rural labour market through brain drain. Using US county-level data, we find a negative relationship between metropolitan patent counts and non-metropolitan labour market performance. Our basic calculation indicates that, between 2005 and 2015, metropolitan technological progress was associated with a relative loss of about 2.5 million non-metropolitan jobs.

KEYWORD:

• knowledge spillovers
• brain drain
• product market competition
• labour market
• urban–rural interrelation

JEL:

• J61
• O33
• R11
• R23

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1. The terms ‘knowledge’ and ‘technology’are used here interchangeably.

2. Lagged patent counts are used to lessen concerns about a simultaneous relation between the unemployment rate and technology.

3. Labour-saving technologies developed in metro areas might disproportionally affect the agriculture and manufacturing sectors in rural regions. In 2015, manufacturing jobs accounted for only 6% of total metropolitan jobs; however, these jobs accounted for 11% of total non-metropolitan jobs (USDA, 2017).

4. Metropolitan technological progress can either create non-metro jobs through commuting or lead to large-scale non-metropolitan-to-metropolitan migration. For details, see Gaile (1980).

5. Bloom et al. (2013) also find that social return (the knowledge spillovers effect) from technological investment dominates private return (i.e., the product market competition effect). Other studies, including Wojan et al. (2018), also observe that innovative firms have a higher survival rate than non-innovators.

6. For the OMB metro/non-metro data set, see the USDA website at https://www.ers.usda.gov/data-products/nonmetropolitan-metropolitan-continuum-codes/.

7. We use USDA 2000 educational attainment data as a substitute for 2004 educational attainment.

8. The US Census Bureau provides data covering all counties in the ACS five-year estimates; these data are collected over a period of five years. The data are only available from 2009.

9. The less-populated non-metropolitan counties that are not in our sample, on average, have about 1.7 patents per year, and about 35% of these counties have zero patents in the entire sample period. On the other hand, non-metro counties in our sample contain, on average, more than 11 patents per year, and fewer than 5% of these counties have zero patents in the entire sample period.

10. Following Jaffe et al. (1993), researchers using patent data commonly assign a patent on the basis of the first-named inventor’s residence. Our data source is the USPTO, which also follows this common practice. For this data set, see https://www.uspto.gov/web/offices/ac/ido/oeip/taf/countyall/usa_county_gd.htm.

11. Following Griffith and Simpson (2004), all stocks containing negative values for any year are set to zero. To avoid the issue of a reference year, we use the average patent counts between the two years as a base to calculate the annual patent growth rate. That is: $g_{k,\left(t,t+1\right)}=(\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}{\mathrm{t}}_{k,t+1}-\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}{\mathrm{t}}_{k,t})/0.5(\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}{\mathrm{t}}_{k,t}+\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}{\mathrm{t}}_{k,t+1})$and ${\overline{g}}_k=1/5{\sum }_{2000}^{2005}{g}_{k,\left(t,t+1\right)}$. If a county is not granted any patents in year t and t + 1, we set $g_{k,\left(t,t+1\right)}=0$.

12. We do not expect the chosen depreciation rate to affect our findings because we control for county and time FE in all regression models.

13. Using data from 469 cities for the period 2011–14, the correlation between the natural log of granted patents and the natural log of total expenditures on R&D lagged one period at the city level equals 0.57. The data on total R&D expenditure at the city level can be found on the SAGE Stats website (http://data.sagepub.com.proxy.lib.ohio-state.edu/sagestats/15926). For the data for total number of patents granted at the city level, see http://data.sagepub.com.proxy.lib.ohio-state.edu/sagestats/14121.

14. Our analysis of the 2007–14 data from 377 counties shows that the correlation between patent applications and granted patent counts is 0.97. For the SAGE Stats data, see https://data.sagepub.com/sagestats/.

15. Giving all patents an equal weight, as we do, understates the role of the greatest innovations. While citations are often used as a proxy for patent quality, they also have their limitations. We leave this issue of quality-adjusted technological stocks for future study.

16. We use the Census 2000 Summary File 1.

17. A correlation table is available from the authors upon request.

18. These studies are conducted at the country level.

19. Following Keller (2002), we also use an exponential decay function instead of the basic inverse distance function as a weighted matrix in a separate analysis. We then conducted the same analysis as those presented in our main empirical analysis (with the unemployment rate as the dependent variable). Despite being statistically insignificant in some estimations, the correlations between $M\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}{\mathrm{k}}_{it}$ and the rural unemployment rate range from +0.12 to +5.34. For this study, these results show that the exponential-decay function is less robust than the basic inverse distance function.

20. Crescenzi et al. (2007) find evidence that social and economic conditions can affect the technological growth rate. We use educational attainment, FE and BB estimators to control for these conditions.

21. When lagged differences are used as instruments for the level equation, the first-stage F-statistic is 31.0. When lagged levels of the dependent variable are used as instruments for the difference equation, the first-stage F-statistic is 22.7.

22. In column (3) of Table 2, the coefficient of the first lag of the log unemployment rate is 0.86 and is significant at the conventional levels. In column (4), the coefficient of the first lag is 0.83 and is significant at conventional levels. The second-lag coefficient is –0.08 and is significant at the 5% level. The results from both columns suggest that the effect of metropolitan technology on the non-metro unemployment rate is greater in the long term than in the short term – consistent with some of the unemployed eventually migrating elsewhere for work. In column (5), the coefficients of the first and second lags of the employment/population ratio are 0.87 and –0.09, respectively. They are both significant at the 1% level.

23. To perform a robustness check on the choice of the number of instruments for the preferred estimator (whose results are in column (4) of Table 2), we limit the numbers of instruments to 56 and 45. The coefficients for the variables of interest are almost identical up to two decimal points compared with the baseline results (the number of instruments is 66).

24. In another analysis, we control for deeper lags of the dependent variables, but the results are virtually unchanged.

25. The statistical insignificance results of Ostock, Rstock and Educ could be due to the short time span of the data and multicollinearity.

26. Evaluating at the sample mean of total labour force: $\begin{array}{rl}& \mathrm{\Delta }\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{U}\mathrm{n}\mathrm{e}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{e}\mathrm{d}\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}\mathrm{s}\\ & =\mathrm{e}\mathrm{x}\mathrm{p}[\frac{\mathrm{\partial }\log (\mathrm{U}\mathrm{n}\mathrm{e}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{e})}{\mathrm{\partial }\log (\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{K}\mathrm{n}\mathrm{o}\mathrm{w}\mathrm{l}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{e}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k})}\\ & \mathrm{l}\mathrm{n}(\mathrm{\Delta }\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{y}\left)\right]{\mu }_{\mathrm{l}\mathrm{a}\mathrm{b}\mathrm{o}\mathrm{r}},\end{array}$where ${\mu }_{\mathrm{l}\mathrm{a}\mathrm{b}\mathrm{o}\mathrm{r}}=47,374$.

27. However, in the regression of the employment/population ratio, the BB estimator rejects the null hypothesis of the Sargan test. To estimate the effect of metropolitan technology on the employment/population ratio, we also use other estimators, including FE and the FD, and these yield similar results. The coefficients of metropolitan technology in both estimations (i.e., FE and FD) are approximately 0.2, but they are not statistically significant at conventional levels.

28. Evaluating at the sample mean of population: $\begin{array}{rl}& \mathrm{\Delta }\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{E}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{e}\mathrm{d}\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}\mathrm{s}\\ & =\mathrm{e}\mathrm{x}\mathrm{p}[\frac{\mathrm{\partial }\log \left(\mathrm{E}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}/\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\right)}{\mathrm{\partial }\log (\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{K}\mathrm{n}\mathrm{o}\mathrm{w}\mathrm{l}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{e}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k})}\\ & \mathrm{l}\mathrm{n}(\mathrm{\Delta }\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{y}\left)\right]{\mu }_{\mathrm{p}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}},\end{array}$where ${\mu }_{\mathrm{p}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}=92,266$.

29. Using the available sample, the:

Predicted number of job $\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{s}=\sum _t\sum _i\left\{\mathrm{e}\mathrm{x}\mathrm{p}\left[{\displaystyle \frac{\mathrm{\partial }\log \left(\mathrm{U}\mathrm{n}\mathrm{e}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{e}\right)}{\mathrm{\partial }\log \left(\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{K}\mathrm{n}\mathrm{o}\mathrm{w}\mathrm{l}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{e}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}\right)}\mathrm{l}\mathrm{n}{\left(\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{y}\right)}_{it}}\right]\right(\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}{)}_{it}\}\approx 2.7\times {10}^6,$

where $i$ denotes a county; t denotes a year; and: $\mathrm{\partial }\log \left(\mathrm{U}\mathrm{n}\mathrm{e}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{e}\right)/\mathrm{\partial }\log \left(\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{K}\mathrm{n}\mathrm{o}\mathrm{w}\mathrm{l}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{e}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}\right)=\mathrm{0.21.}$Similarly, the predicted number of job losses can also be calculated as follows:

$\mathrm{N}\mathrm{u}\mathrm{m}\mathrm{b}\mathrm{e}\mathrm{r}\mathrm{o}\mathrm{f}\mathrm{J}\mathrm{o}\mathrm{b}\mathrm{L}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{s}=\sum _t\sum _i\left\{\mathrm{e}\mathrm{x}\mathrm{p}\left[{\displaystyle \frac{\mathrm{\partial }\log \left(\mathrm{E}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}/\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\right)}{\mathrm{\partial }\log \left(\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{K}\mathrm{n}\mathrm{o}\mathrm{w}\mathrm{l}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{e}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}\right)}\mathrm{l}\mathrm{n}{\left(\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{y}\right)}_{it}}\right]\right(\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}{)}_{it}\left\}\right|\approx 2.1\times {10}^6,$

where:

$\mathrm{\partial }\log \left(\mathrm{E}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}/\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\right)/\mathrm{\partial }\log \left(\mathrm{U}\mathrm{r}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{K}\mathrm{n}\mathrm{o}\mathrm{w}\mathrm{l}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{e}\mathrm{S}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}\right)=-0.05$.

30. Wages and Income (CA1) data are from the US Bureau of Economic Analysis (https://www.bea.gov/itable/iTable.cfm?ReqID=70&step=1#reqid=70&step=1&isuri=1).

31. The base years used to calculate the consumer price index are 1982–84. The index was retrieved on 6 November 2017 from the Federal Reserve Bank of Minneapolis (https://www.minneapolisfed.org/community/teaching-aids/cpi-calculator-information/consumer-price-index-and-inflation-rates-1913).

32. It is unlikely that there is a reverse causality between a small non-metropolitan economy and state-level GDP.

33. For both the wage and income estimations, the FD estimator reports coefficients of the metropolitan technological stock that resemble those estimated by the FE estimator. We also employ the BB estimator using one lag of the dependent variables. Using the BB estimator for wage estimation, the coefficient of the metropolitan stock variable is –0.40 and is significant at the 5% level. In the case of income estimation, the BB estimator yields a coefficient of the metropolitan technological stock of −0.23; however, it is insignificant at conventional levels.