1. Introduction
Recent extreme heat waves attributed to global warming have sparked an interest in the question whether society will be able to invent new technology to adapt to the likely greater frequency and severity of such shocks (Dechezlepretre et al. 2020). In this regard it may be reassuring that human history is full of examples of the development of technologies that have allowed mankind to reduce the impact of extreme climate and weather events, such as the development of irrigation systems (6000 BC; Sojka et al. 2002), the construction of levees (1000 BC; Butzer 1976), or the invention of the lightning rod by Benjamin Franklin (1752; Cohen 1952). Nevertheless, we need to understand what is likely to drive such innovation in order to create the most conducive regulatory environment for it.
Zilberman et al. (2012) note that while technological adaption to climate may in principle be reactive or proactive, most adaptation activities in the past appeared to have been of the latter form.1 Relatedly, in a more recent context, Dechezlepretre et al. (2020) show using global patent data that although innovation related to climate adaptation has risen considerably in parallel with the growing public concern over climate change and its implications, this increase is no greater than the rise in non-climate-related innovative activity.
Empirically discerning whether weather-shock-mitigating innovation has been proactive is a difficult task, as it would require linking such innovation to a measure of expectations of future changes in the distribution of climate shocks. In contrast, there are already a number of studies that have explicitly tested whether past extreme weather events can induce technological adaptation (i.e., whether innovation may have been reactive). More precisely, based on panel data of 28 countries over a period of 25 years, Miao and Popp (2014), in a first study addressing the issue, find that the occurrence of droughts and floods increases the number of risk-mitigating patents.2 A similar positive effect was estimated by Li (2017) for the United States and by Hu et al. (2018) for regions in both modern and premodern China. Looking specifically the Dust Bowl effects in the United States in the 1930s, Moscona et al. (2021) show that the more crops were exposed to soil erosion, the more new varieties were introduced. Noy and Strobl (2021) link hurricanes and related patents and similarly find a boost in damage-mitigating patents after a storm. Yet, in contrast with this literature that focuses on the short-run reaction, they find that in the long run hurricane-related innovation actually declined.
In this paper we re-examine the possible reactive nature of innovation to extreme climate events by focusing on the impact of extreme heat on innovation in air conditioning technology in the United States. There are a number of reasons why a focus on heat and air conditioning is an interesting context within which to study the issue. First, extreme heat is possibly the main driver of climate-related mortality risk in high-income countries, and air conditioning has been shown to have been an important technology in reducing this risk (Barreca et al. 2016). Second, air conditioning technology has also been a very important factor in shaping the rapid industrialization and economic development that has characterized the period 1920–70 in North America in particular, and later elsewhere around the world (Cooper 1998). Indeed, the first building to have been built with an air conditioning system was, appropriately, the New York Stock Exchange in Wall Street, built in 1902.3
The United States has been the primary location of the development of this technology and where patents have been issued since the very beginning of the twentieth century, shaping the evolution of this technology globally. For example, as early as 1904, 1906, and 1907, patents were issued for various humidity-regulating devices (Cooper 1998), and these were then followed with further patented innovations throughout the century.4 Finally, cooling technology may possibly be the most important technological innovation that will allow adaptation to climate change, especially in hotter and poorer countries where air conditioning is not yet so common.5 Ultimately, in some places, life in the future may no longer be possible without it (IPCC 2018).
To conduct the analysis in this paper we combine two rich datasets. On the one hand, we use historical geo-located patent data, where we identify those innovations that are relevant specifically to air conditioning. We identify the county in which one or more inventors resides, and then couple these with gridded climate data that allows us to construct a heat wave index that uses both temperature and relative humidity jointly to measure thresholds of the comfort level of the human body. This provides us with an annual county-level panel dataset for the United States covering nearly 100 years (1919–2011).
Our analysis is conducted with this county-level data for four reasons. First, we are specifically testing for the importance of local personal experience in motivating and determining innovation activity, and county-level analysis is the smallest geo-administrative unit in which our data can be identified. Second, our methodological approach (detailed below) allows us to account for any nationwide changes (including nationally news-worthy heat waves) that may affect local innovation activity. Third, econometric testing concludes that there are no spatial affects that make heat waves in one location correlate with innovation activity in nearby locations. Fourth, even if there are national or regional effects of “nearby” heat waves, our empirical approach means that these will only attenuate any effect we identify. One can thus interpret our findings as a lower bound on what may be the true global effect of heat waves on innovation in air conditioning.
Ultimately, our analysis shows that there is indeed a short-term 2-yr boost in inventions in air conditioning technology after a heat wave, where the average extreme heat exposure has increased air conditioning–related innovation by about 7.5%.
The remainder of the paper is organized as follows. A detailed description of our data and method is provided in section 2. The results of our econometric estimations, and some examination of their robustness for the determinants of air conditioning innovation, are described in section 3. Section 4 concludes.
2. Data and methods
a. Heat index
To create a heat index we rely on the U.S. National Weather Service heat index chart (NWSHIC).6 This heat index is a measure of the human body’s comfort level based on the combined effect of temperature and (relative) humidity, as modeled by Steadman (1984). To this end the NWSHIC categorizes combinations of these two aspects into four categories (see Fig. 1) that determine the likelihood of heat disorders with prolonged exposure or strenuous activity: (i) caution, (ii) extreme caution, (iii) danger, and (iv) extreme danger.
National Weather Service heat index.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0095.1
b. Population exposure data
To derive historical exposure to heat waves within counties, we use the gridded decadal population estimates from the HYDE dataset version 3.2.1, which provides global population maps at a spatial resolution of 0.5 arc-min for every 10 years from 1700 until 2000, and yearly thereafter (Klein Goldewijk et al. 2017). For the period prior to 2000, we linearly interpolate population figures to arrive at annual values.8
c. Climatic data
We use two sources of climatic data. The first is the Livneh daily continental U.S. (CONUS) near-surface gridded daily data, which are derived from meteorological data of approximately 20 000 NCDC stations gridded to a 1/16° spatial resolution, running from 1 January 1915 to 31 December 2011. For the purpose here we use the measures of daily maximum and minimum temperature and precipitation. Annual averages of precipitation and daily temperature are used as the climate control variables in Cit. We also use the daily maximum temperature values to estimate the saturation water vapor pressure in Eq. (3).
Since the CONUS data do not provide measures of vapor pressure deficit needed for Eq. (2), we resort to the PRISM database, where in situ point measurements are ingested into the Parameter elevation Regression on Independent Slopes Model (PRISM) statistical mapping system, to generate a number of hydrometeorological variables at 4-km resolution from 1895 onward. One may want to note that the monthly PRISM data are used as a scalar to construct the variables in Livneh (Livneh et al. 2015). We use the minimum average values of the vapor pressure deficit, rather than the average or maximum, as these provided the most conservative estimates of heat index days. Given that we want to identify the daily occurrences of the heat index categories we assume that minimum vapor pressure deficit is constant through a month. As both the CONUS and the PRISM data are at a higher spatial resolution than the HYDE population data, we calculate and use average values of all climate variables at their given frequency within the HYDE cells.
d. Patent data
To construct county-level time-varying patents, we collect data from two sources. From 1976 onward, we use information from the United States Patent and Trademark Office’s PatentsView database (https://patentsview.org/), which contains information on all patent activity in the United States. In particular we use this data to determine the filing year of granted patents and the geographic location (latitude and longitude) of the inventor(s). We limit our analysis to those U.S. patents for which at least one of the inventors was located in the United States, and allocate their location to the appropriate county. Prior to 1975 we use the county-level location attributed to U.S.-located inventors as constructed by Petralia et al. (2016). More specifically, Petralia et al. (2016) develop and implement a procedure using the digitized records of original patent documents that allows them to recover the county of the inventor of the patents, rather than the assignee (if they were different). In using this information we note that where there was more than one inventor for any patent, we attributed the patent to the counties of all inventors.9
To identify air conditioning–related inventions we searched each patent’s title and text for some combination of the occurrence of the terms air conditioner or air conditioning that indicated that the patent was related to an air conditioner or its use in some fundamental way. It was important in this regard that the two words—1) air and 2) conditioner/conditioning—were mentioned sequentially as many patent descriptions employed these terms in different contexts. One exception that we did allow for was the use of air cooler and conditioning and air cooling and conditioning as there were several patents that described inventions that involved both air cooling and conditioning. One should note in this respect that we focus in this study on air conditioning technology only, or as a combination with evaporation cooling technology, and not patents that solely involve the latter. This is done because evaporative cooling is fundamentally different than air conditioning in that it simply imports exterior air from outdoors and passes it through a wet medium, and thus is limited in its ability to cool by the existent level of humidity, making it unsuitable for humid climates.
The search approach just described resulted in a total of 101 012 patents for the sample period (1915–2011). Of these, 30 610 were captured by the term air conditioner, 70 346 by air conditioning, 5 by air cooler and conditioner, and 51 by air cooling and conditioning. A more detailed analysis of about 1000 random patents identified accordingly showed that incidences of combination these words accurately highlighted patents that were directly associated with heat-related cooling systems, as all but three were clearly related to air conditioning technology and not to heating. Thus one can be reasonably confident that our classification procedure allows us to categorize most of the universe of patents as either air conditioning–related or not if these aforementioned terms were used. One may want to note that for about two-thirds of the identified patents we were able to link the sectoral classification code of Marco et al. (2015), allowing us to identify the broad sector the patent falls in. These patents are assigned standard Cooperative Patent Classification (CPC) codes. Finally, the data also allow us to assign the patents to a company if ownership is not with the inventor but at the company level. This is the case in 84.4% of the patents.
e. Regression methods
Hubert/White/sandwich robust standard errors are calculated. The use of PCFE with robust standard errors provides a number of advantages. First, it produces the same coefficients, as well as associated covariance matrix, as the unconditional Poisson fixed effects estimator, and hence does not suffer from the incidental, typical of many nonlinear estimators (Cameron and Trivedi 2013). Additionally, it is robust to misspecification, such as lack of equidispersion, an excessive number of zeros, and dependence over time, as long as the conditional mean is correctly specified (Wooldridge 1999). As shown by Bertanha and Moser (2016) spatial dependence may change the variance of the PCFE estimator if this dependence is time varying. To test for time-varying spatial dependence, we employ the test developed by Bertanha and Moser (2016).
3. Results
a. Data sample and summary statistics
Our sample consists of a balanced panel of annual patent activity, heat index days, and climate controls over 97 years (1915–2011) covering 3084 counties in the coterminous United States. Given that we experiment with lagged effects of up to 4 years, we limit the sample period to 1919–2011. One should also note that since over 99% of heat index positive days took place after March in any year, we redefine years as running from 1 March of any year to the end of February of the following year to minimize the likelihood that we are not attributing patents filed before March to the heating index positive days that are likely to have occurred after their filing date.
Summary statistics for all our variables are provided in Table 1. Accordingly, 0.19 patents per county, on average, related to air conditioners are filed every year, but with considerable variation. In contrast, general patents for innovations not related to air conditioners are filed about 25 times a year per county, although there is large variability across time and space. One may note that the largest share of the classified patents was in the Others–Miscellaneous sector (27.6%), followed by Motors, Engines, and Parts (8.6%), Heating (8.6%), Power Systems (6.7%), Computer Hardware and Software (6.4%), Mechanical–Miscellaneous (6.1%), Chemicals–Miscellaneous (6.0%), Communications (4.0%), and Transportation (3.4%). Looking at the distribution of CPC codes one finds that the top 10 categories are Refrigeration (18.1%), Heat Exchange (6.7%), Ventilation (4.5%), Automatic Temperature and Humidity Regulation (2.7%), Pumps (2.5%), Internal Combustion Engines (2.3%), Data Processing–Vehicles, Navigation, and Other Location Systems (2.2%), Communications–Electrical (2.0%), Static Structures (1.9%), and Data Processing–Generic Control Systems and Specific Apparatus (1.8%).
Summary statistics (AC = air conditioning; HI = heat index).
Regression results for AC patents. Robust standard errors are given in parentheses. One and two asterisks indicate 5% and 1% significance levels, respectively. All specifications include year dummies and state-specific time trends, as well as appropriately lagged precipitation and temperature controls. ED = extreme danger category.
In terms of ownership of the patents by companies, the inventor could assign ownership to a company. However, the data show no great concentration of patents in specific companies. The top five corporations were owners of less than 10% of patents. These are Denso Co. (2.0%), Carrier Co. (1.9%), General Electric (1.4%), Toyota (1.3%), and LG (1.2%). Thus, out of these top five patent owners, only two are headquartered in the United States (Carrier in Florida and General Electric in Massachusetts).
If one defines the heat index in terms of those days that are identified as extreme danger, there were on average 15 days yr−1 over our sample period, where the highest observed value was 201 days (Highlands County, Florida, in 1986). If one lowers the threshold to that of danger status, the average number of days more than doubles to 32, while a classification of extreme caution indicates an average of 45 days yr−1. Finally in terms of the climatic controls the average county-level monthly precipitation was nearly 80 mm, whereas the annual average daily temperature fluctuates considerably around its 12°C mean.
We depict the logged total number of air conditioner related patents (+1) as well as the average annual number of extreme danger heat index days over our sample period in Fig. 2. As can be seen, the average number of air conditioning–related patents filed per year has increased substantially since 1919, with a particularly stark rise since the mid-1980s. However, there are also clear fluctuations from year to year. Examining the evolution of the average county level number of days that reached extreme danger status indicates no real trend over time, but a large variability. Moreover, it is difficult to visually identify any comovement between the two series, where the raw correlation was −0.015.
Average heat wave index (right axis; red) and total air conditioning (AC) patent counts (left axis; blue).
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0095.1
We map the log of the average county level total number of air conditioner–related patents (+1) in Fig. 3. There is a high concentration of air conditioning–related innovation in Florida, in many counties in the Northeast (north of Virginia), and near the Great Lakes.11 Additionally, one can see a relatively higher number of patents in California and Arizona, as well as in parts of Texas. Comparing this distribution to that of other patents in Fig. 4 suggests that at least some of the unequal geographic distribution may be due to generally greater innovative activities in these areas.
Log total number of AC patents (+1) [1919–2011].
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0095.1
Log total number of non-AC patents (+1) [1919–2011].
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0095.1
Figure 5 depicts the average annual days that surpass the heat index classification of extreme danger. Accordingly, and unsurprisingly given the general climatic conditions, it is particularly counties in Florida and Georgia and the coastal states to the west up to and including Texas that experienced the highest number of days at which the temperature and relative humidity combination reaches these uncomfortable levels.
Average heat index [1919–2011].
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0095.1
b. Main regression results
We first investigated whether there is time-varying spatial correlation, potentially undermining the validity of the PCFE estimator, present in Eq. (4) using the test developed by Bertanha and Moser (2016). The resultant test statistic (1.793) on the null hypothesis that the spatial correlation is time invariant indicated that this is not the case.
Coefficient estimates of HI from Eq. (4), starting from allowing only for a contemporaneous HI and the increasingly including additional lags, are shown in Table 2. One should note that out of the 3087 counties with some patent activity over our sample period, only 1746 had at least one patent pertaining to air conditioning, leaving this, after controlling for county fixed effects, to be our estimation sample. As can be seen, there is no contemporaneous effect of HI on local patent filing activity related to innovation in air conditioning. However, as one increases the number of lags for HI a different picture emerges. More precisely, one year after at least some days of sufficient temperature and humidity to be classified as extremely dangerous, the number of patents filed that relate to air conditioning rises and this effect persists up to two years after the heat wave has occurred. After three years, the significant impact disappears. Taken at face value, the estimated coefficients suggest that the average number of extreme danger heat days a county experiences (≈15) causes air conditioning patent filings to increase by 3% one year after and 4.5% two years after the event. If one takes the most extreme heat index observation in our sample period (≈200), then the corresponding increases are 40% and 60%, respectively.
Regression results: Different thresholds, non-AC patents, and sample periods. Robust standard errors are given in parentheses. One and two asterisks indicate 5% and 1% significance levels, respectively. All specifications include year dummies and state-specific time trends, as well as appropriately lagged precipitation and temperature controls. ED = extreme danger; D = danger; EC = extreme caution.
c. Robustness checks
Up until now, we have only considered days that fall within the extreme danger (ED) category. In the first column of Table 3 we lower the threshold to also include days that are only classified as danger (D). Accordingly, this renders the coefficient on HIit−1 insignificant. While heat index days two years earlier still increases air conditioning patent-filing activity significantly, the coefficient drops from 0.003 to 0.002. In the second column we further reduce the threshold to include days that are classified as extreme caution (EC). As can be seen, the coefficient on HIit−1 is now statistically insignificant, although it flips its sign. In contrast, one still finds a patent activity boosting impact two years after positive heating index days, although as with the D specification, the coefficient is lower than for the highest threshold.
Regression results: further robustness checks. Robust standard errors are given in parentheses. One and two asterisks indicate 5% and 1% significance levels, respectively. All specifications include year dummies and state-specific time trends except for values in parentheses, which only includes state-specific time trends. All specifications include appropriately lagged precipitation and temperature controls. The ED threshold of heat index is used in all but the last column.
To ensure that high combinations of temperature and humidity do not just increase patent activity in general, we re-estimated Eq. (4) for all non–air conditioning patent-filing activity in the third column of Table 3. The results indicate that extreme danger heating days have no effect for other types of patents. Since this estimation involves a larger sample size, because there were many counties where other patents were filed over our sample period, we also reran the regression only with counties that had at least one air conditioning patent filed. As can be seen from the last column, the results are qualitatively identical to the full sample; extreme heat days have no statistical association with other types of patenting activity.
Given that air conditioning was not widely adopted until the second half of the twentieth century (Biddle 2008), we divided the data into the period before and after 1950 and estimated our main specification for these two samples separately. As can be seen in the last two columns, for the earlier period the impact is more delayed than for the full sample, where a significant coefficient is only found 3 years later. In this case, however, the estimate is twice as large as the combined effects of t − 1 and t − 2 for the full sample. In contrast, for the period after 1950 the results are similar to the full sample. Thus, extreme heat events appear to lead to air conditioning innovation more quickly for the latter period, but the overall magnitude of the impact is smaller.
We also explored whether there are differences across regions with different extreme heat experience. More specifically, we divided the sample into those counties above the mean HI (14.739) and those below, and reran our main specification for these two subsamples separately, shown in the first and second columns of Table 4. Accordingly, for the counties that have above-average exposure the results are qualitatively similar to the full sample, but the effects are quantitatively higher. In contrast, while for low-exposure counties the impact at t − 2 is similar to that for their high counterparts, there is no effect at t − 1. This suggests that perhaps in low-exposure counties an extreme heat day is viewed as a rarity and hence less likely to induce immediate response in terms of innovation to mitigate such episodes.
To ensure that the estimated impact of extreme heat days is not picking up general increases in patent activity, in the third column of Table 4 we included appropriately lagged total non–air conditioning patent activity as additional regressors. As can be seen, this only marginally reduced the quantitative impact of t − 2 extreme heat incidences.
One may want to note that the inclusion of year dummies in all our specifications captures all common macroeconomic shocks, including extreme heat episodes elsewhere that may have affected national actual and perceived demand for air conditioning technology. Under the assumption that there are no other confounding aggregate factors that are correlated with local extreme heat days and local air conditioning innovation, one can also examine the role of extreme heat outside of the county. To this end we reran the main specification excluding yearly dummies, and instead included a nationally population weighted extreme heat index (NHI). From the results depicted in the fourth column, the local estimates remain essentially unaltered. Examining the national level index, one finds that there are positive impacts on local innovative activity from t − 2 until t − 4. There is a negative significant impact of national extreme heat days at t − 1.
Arguably, an important component of the heat index employed here is that it not only takes account of days of high temperature but the combination of these with high humidity, which together interact to adversely affect the comfort level of the human body. To assess whether the latter is indeed a necessary component to encourage local innovation in air conditioning technology, we alternatively created a population index of the number of days above 90°F (∼32°C), where the choice of this threshold coincides with lowest value of temperature that is necessary to induce the classification of extreme heat if humidity is high enough; see Fig. 1. However, using a temperature-only-based classification of an extreme heat event does not induce any significant impact of air conditioning innovation, as shown in the last column of Table 4.
Finally, as an additional placebo test we included up to four leads of the HI index in the main specification. The results of the estimated coefficients along with their 95% confidence intervals are shown in Fig. 6. Accordingly, while the lagged effects at t − 1 and t − 2 persist, there are no significant impacts of the leads of HI.
Placebo test of timing.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0095.1
d. Measurement error
There are a number of sources of possible measurement errors in our econometric analysis that deserve further discussion. Most obviously, the procedure of classifying patents for air conditioning technology will involve some error. While a random sample check indicated that the problem of misclassifying patents as related to air conditioning when they were in fact not was not significant, we may also have missed patents that involved technologies important for air conditioning but that did not incorporate our search terms in their description of the patent. If this was substantial, it would have increased the imprecision of our estimates, possibly leading to insignificant findings where there was an actual impact. Similarly, there may be some imprecision due to faulty geographic classifications of patents, issues in the method developed by Petralia et al. (2016), our decision to allocate the patent equally across counties if there were multiple inventors, and the possibility that a patent may have been filed in a county different from where the innovation took place. All of these should have also only attenuated our findings.
Perhaps most importantly, our measure of innovation (patents) may be fairly limited in its ability to capture all innovation related to air conditioning technology. One worry may be that patents are not a good proxy for local innovation in some domains, even if they are frequently used as such (Acs et al. 2002). It is a decision of the innovator, or their employer, to file for patent protection, and it may be the case that they choose not to do so. In other cases, the innovation might not be patentable. Ultimately, patent protection is sought only if the new technology cannot be kept secret, or if it involves tacit knowledge that is not patentable. Neither of these conditions likely apply in the case of air conditioning, so we think these concerns are muted.
Another concern is about the quality of the innovations being developed. Many if not most patents are not commercially successful or are a technological dead end. The distribution of citations of patents, for example, tends to be highly skewed, with few patents getting much attention and many languishing in obscurity (Kuhn et al. 2020).12 This does not introduce any bias into our estimates (we can still identify the impact of heat waves on air conditioning patents) but if the measurement error this skewness introduces is correlated with extreme heat events, then this could possibly introduce systematic bias in our interpretation of the impact of extreme heat events on the extent of meaningful and consequential innovation (rather than “just” patents). If during heat waves the quality of innovations rose more (less) than the number of patents, then the estimates would be downward (upward) biased. However, we see no reason to expect such correlation.
Inherently the heat index HI employed here will be measured with some error as well, especially in terms of capturing true and impactful heat waves. First, once an event satisfies the criterion of crossing above the two-dimensional threshold for at least two consecutive days, the marginal impact of each day is assumed to be constant, which may not reflect reality. Second, there is no scientific consensus on the definition of a heat wave (Robinson 2001), where this may vary geographically, or even culturally. These weaknesses of the index employed here are also likely to introduce an attenuation bias into our regression analysis, rather than cause us to wrongly identify a statistical relationship.
4. Conclusions
In this paper we investigated whether the occurrence of extreme heat events (heat waves) generates innovation in cooling technology, as measured by patents, using a 90-yr county-level panel dataset for the United States. We find that in the 2 years after a heat wave the number of air conditioning patents that are issued increases. We find no such increase in the frequency of general patents, and conclude that the heat wave itself is the reason for the increased innovation activity in cooling technology.
Importantly, as Barreca et al. (2016) have shown in their study of the temperature–mortality relationship in the twentieth-century United States, it was the spreading of air conditioning technology that led to the dramatic reduction of heat wave–induced mortality in the United States. This risk, however, has not been eliminated. Anthropogenic climate change is already causing more heat waves, and heat waves that are more intense than the ones experienced in our recorded history, as was evidenced in the unprecedented heat wave in the U.S. Northwest in the summer of 2021 (Philip et al. 2021). Extreme heat waves are going to become even more prevalent under climate change, at least in some places (Dosio et al. 2018). It is therefore inevitable, given also the greenhouse gas emissions reductions that are necessary, that we will continue to require further improvements and more innovations in air conditioning technology.
Our findings suggest that some new innovation will happen, but our investigation also suggests that any impact is fairly short lived. It is possible, and indeed plausible, that since the pace of climatic change is accelerating, the reactive innovation that our model predicts will not suffice to achieve both the mitigation and adaptation goals we need to reach. It is also noteworthy, in this context, to note that we find no evidence for proactive innovation that is undertaken in anticipation of these future risks.
Given the already widespread use of air conditioning in the United States right now, we also note that the two main areas in which new innovation is needed are 1) to make air conditioning technology more affordable, so that it can also be deployed on large scale in low-income countries, and 2) to reduce the environmental footprint of this technology, specifically to reduce the amount of greenhouse gas emissions it produces. These two aims are unlikely to be motivated by the heat wave exposure channel we describe. Most innovation in air conditioning has historically occurred in the United States, where the risk today is dramatically reduced when compared to its early days (and when our empirical analysis begins) in the early twentieth century. We also note that more innovation today, at least in the U.S. context, is on heat pump technology, one that combines both heating and cooling. But, again, this is not going to deal with the main concerns of heat in low-income countries (mostly in the tropics) that do not require this additional heating ability.
On a cautionary note one should emphasize that our findings do not necessarily imply that the incentives to spend large resources on innovation to reduce costs, and reduce emissions, are now in place in this cradle of air conditioning. Rather our analysis has shown that innovators respond to their personal experience in developing new ameliorating technology to deal with heat wave events. It plausible that this phenomenon should be taken into account when designing incentives and support systems for innovation—for example, emphasizing building the infrastructure (physical, financial, or intellectual) to support this demand for reactive innovation. In this regard, there is unfortunately not much that is known about the mechanisms behind this reduced-form link between disasters (heat waves or other extreme events) and innovation in disaster mitigation technology. It is thus more difficult to come up with very concrete policy proposals based purely on the reduced-form links highlighted here. This should constitute an important field of future research.
REFERENCES
Ackermann, M., 2002: Cool Comfort: America’s Romance with Air-Conditioning. Smithsonian Institution Press, 214 pp.
Acs, Z. J., L. Anselin, and A. Varga, 2002: Patents and innovation counts as measures of regional production of new knowledge. Res. Policy, 31, 1069–1085, https://doi.org/10.1016/S0048-7333(01)00184-6.
Barreca, A., K. Clay, O. Deschenes, M. Greenstone, and J. S. Shapiro, 2016: Adapting to climate change: The remarkable decline in the US temperature–mortality relationship over the twentieth century. J. Polit. Econ., 124, 105–159, https://doi.org/10.1086/684582.
Bertanha, M., and P. Moser, 2016: Spatial errors in count data regressions. J. Econ. Methodol., 5, 49–69, https://doi.org/10.1515/jem-2014-0015.
Biddle, J., 2008: Explaining the spread of residential air conditioning, 1955–1980. Explor. Econ. Hist., 45, 402–423, https://doi.org/10.1016/j.eeh.2008.02.004.
Butzer, K. W., 1976: Early Hydraulic Civilization in Egypt: A Study in Cultural Ecology. University of Chicago Press, 134 pp.
Cameron, A. C., and P. K. Trivedi, 2013: Regression Analysis of Count Data. 2nd ed. Cambridge University Press, 566 pp.
Cohen, I. B., 1952: The two hundredth anniversary of Benjamin Franklin’s two lightning experiments and the introduction of the lightning rod. Proc. Amer. Philos. Soc., 96, 331–366.
Cooper, G., 1998: Air-Conditioning America: Engineers and the Controlled Environment, 1900–1960. John Hopkins University Press, 229.
Dechezlepretre, A., S. Fankhauser, M. Glachant, J. Stoever, and S. Touboul, 2020: Invention and global diffusion of technologies for climate change adaptation: A patent analysis. World Bank, 55, https://openknowledge.worldbank.org/handle/10986/33883.
Dosio, A., L. Mentaschi, E. M. Fischer, and K. Wyser, 2018: Extreme heat waves under 1.5°C and 2°C global warming. Environ. Res. Lett., 13, 054006, https://doi.org/10.1088/1748-9326/aab827.
Eccel, E., 2012: Estimating air humidity from temperature and precipitation measures for modelling applications. Meteor. Appl., 19, 118–128, https://doi.org/10.1002/met.258.
Hu, H., T. Lei, J. Hu, S. Zhang, and P. Kavan, 2018: Disaster-mitigating and general innovative responses to climate disasters: Evidence from modern and historical China. Int. J. Disaster Risk Reduct., 28, 664–673, https://doi.org/10.1016/j.ijdrr.2018.01.022.
IPCC, 2018: Special Report on Global Warming of 1.5°C. Intergovernmental Panel on Climate Change, 630 pp.
Klein Goldewijk, K., A. Beusen, J. Doelman, and E. Stehfest, 2017: Anthropogenic land use estimates for the Holocene—HYDE 3.2. Earth Syst. Sci. Data, 9, 927–953, https://doi.org/10.5194/essd-9-927-2017.
Kuhn, J., K. Younge, and A. Marco, 2020: Patent citations reexamined. RAND J. Econ., 51, 109–132, https://doi.org/10.1111/1756-2171.12307.
Li, H., 2017: Innovation as adaptation to natural disasters. Tech. Rep., University of Waterloo, 47 pp., https://uwaterloo.ca/economics/sites/ca.economics/files/uploads/files/disaster_innovation.pdf.
Livneh, B., and Coauthors, 2015: A spatially comprehensive, hydrometeorological data set for Mexico, the U.S., and southern Canada 1950–2013. Sci. Data, 2, 150042, https://doi.org/10.1038/sdata.2015.42.
Marco, A. C., M. Carley, S. Jackson, and A. Myers, 2015: The USPTO historical patent data files: Two centuries of innovation. SSRN Electron. J., https://doi.org/10.2139/ssrn.2849514.
Miao, Q., and D. Popp, 2014: Necessity as the mother of invention: Innovative responses to natural disasters. J. Environ. Econ. Manage., 68, 280–295, https://doi.org/10.1016/j.jeem.2014.06.003.
Monteith, J., and M. Unsworth, 2013: Principles of Environmental Physics: Plants, Animals, and the Atmosphere. Academic Press, 403.
Moscona, J., and Coauthors, 2021: Environmental catastrophe and the direction of invention: Evidence from the American Dust Bowl. Tech. Rep., 61 pp., http://economics.mit.edu/files/17642.
Noy, I., and Strobl, E., 2021: Creatively destructive hurricanes: Do disasters spark innovation? Tech. Rep., 26 pp.
Orlove, B., 2005: Human adaptation to climate change: A review of three historical cases and some general perspectives. Environ. Sci. Policy, 8, 589–600, https://doi.org/10.1016/j.envsci.2005.06.009.
Petralia, S., P.-A. Balland, and D. L. Rigby, 2016: Unveiling the geography of historical patents in the United States from 1836 to 1975. Sci. Data, 3, 160074, https://doi.org/10.1038/sdata.2016.74.
Philip, S., and Coauthors, 2021: Rapid attribution analysis of the extraordinary heat wave on the Pacific coast of the US and Canada June 2021. 37 pp., https://www.worldweatherattribution.org/wp-content/uploads/NW-US-extreme-heat-2021-scientific-report-WWA.pdf.
Robinson, P. J., 2001: On the definition of a heat wave. J. Appl. Meteor. Climatol., 40, 762–775, https://doi.org/10.1175/1520-0450(2001)040<0762:OTDOAH>2.0.CO;2.
Sojka, R., D. Bjorneberg, and J. Entry, 2002: Irrigation: An historical perspective. Encyclopedia of Soil Science, 1st ed. R. Lal, Ed., Marcel Dekker Inc., 745–749.
Steadman, R. G., 1984: A universal scale of apparent temperature. J. Appl. Meteor. Climatol., 23, 1674–1687, https://doi.org/10.1175/1520-0450(1984)023<1674:AUSOAT>2.0.CO;2.
Wooldridge, J. M., 1999: Distribution-free estimation of some nonlinear panel data models. J. Econ., 90, 77–97, https://doi.org/10.1016/S0304-4076(98)00033-5.
Zilberman, D., J. Zhao, and A. Heiman, 2012: Adoption versus adaptation, with emphasis on climate change. Annu. Rev. Resour. Econ., 4, 27–53, https://doi.org/10.1146/annurev-resource-083110-115954.
Zilberman et al. (2012) refers to the study by Orlove (2005), who discusses the case studies of the classic Maya of Mexico and Central America, the Viking settlements in Greenland, and the U.S. Dust Bowl.
Miao and Popp (2014) find a similar effect for nonclimatic natural disasters, namely, earthquakes.
The term “air-conditioning” was coined by Stuart Cramer, an engineer, in 1906. The term is derived from the term “yarn-conditioning”—the practice of maintaining the humidity of cotton yarn to facilitate its use in textile production. This hints at the pivotal role of humidity control in shaping the beginning of widespread air conditioning, focusing especially on industrial needs in the textiles and tobacco industries (Cooper 1998).
For example, the technology used in the window (single-room) AC unit was mostly patented in the mid-1930s, but only really took off with mass production after WWII with the growth of the suburbs and the democratization of this technology for the masses (Ackermann 2002; Biddle 2008). By the early 1950s, more than 100 brands of AC units were being marketed, based on various evolving technological patents.
For many years, air conditioning was largely an American obsession (Ackermann 2002).
We note that this is defined at the year prior to t in order to ensure that the population distribution within a county is not affected by the heat index in year t.
One should note that the cell centroids of the population data serve as points mi within counties i for use in Eq. (1).
Simply attributing the patent to the first listed inventor made no qualitative difference, and only very marginal quantitative difference, to our results. Details are available from the authors upon request.
Including such county-level fixed effects also ensures that any accumulation of knowledge is captured by HI, as long as this is merely additive.
Ackermann (2002) also document how much of the development of this technology occurred in the temperate climates of the U.S. Northeast and Midwest.
One possibility to address this in our analysis is to weigh patents by their citations. Unfortunately, linking of citations across patents is limited to only to the very recent period.
