Cell Phones and Brain Cancer: A Twenty-Year Cross-Country Analysis

Hugo M. Mialon and Erik T. Nesson

October 13, 2015

Abstract We empirically investigate the association between cell phones and brain cancer. Using a cross-country panel dataset of brain cancer death rates spanning from 1990 to 2012, we find that mobile phone subscription rates are positively associated with death rates from brain cancer. Specifically, one more mobile phone subscription per 100 people is associated with 0.03 more brain cancer deaths per 100,000 people 15 years later. As a falsification test, we find no positive association between mobile phone subscription rates and deaths from rectal, stomach, breast or lung cancer or ischemic heart disease.

JEL classification: I15, O33 Keywords: Mobile phones, cell phones, brain tumor, brain cancer

* Hugo M. Mialon, Department of Economics, Emory University, Atlanta, GA 30322 ([email protected]); Erik T. Nesson, Department of Economics, Ball State University, Muncie, IN 47306 ([email protected]). We are grateful to Dan Hamermesh and Andrew Francis-Tan for helpful comments.

1. Introduction Although cell phones and information technology have revolutionized society, questions remain about negative externalities arising from increased technology use, including automotive accidents and fatalities (Kolko 2009; Abouk and Adams 2013; Asbridge et al. 2013) and decreased learning outcomes (Kuznekoff and Titsworth 2013; Beland and Murphy 2015; Thornton et al. 2015). Additionally, the International Agency for Research on Cancer (2011) declared cell phones a potential carcinogen, although evidence on the link between cell phones and brain cancer is differing and inconclusive. In this paper, we provide new evidence on the relationship between mobile phone use and brain cancer. We use data on country-level brain cancer death rates between 1990 and 2012 from the World Health Organization and country-level mobile phone subscription rates from the World Bank dating back to 1960. We find a positive and statistically significant relationship between mobile phone subscriptions and brain cancer death rates.

In our preferred

specifications, one more mobile phone subscription per 100 people is associated with 0.03 additional brain cancer deaths per 100,000 people 15 years later, statistically significant at the one percent level. Translated into elasticities, a one percent increase in the number of mobile phone subscriptions per 100 people is associated with a 0.008 percent increase in brain cancer death rates 15 years later. These results are fairly robust to a number of specification tests, and we run a number of falsification tests where we examine the relationship between mobile phone subscription rates and mortality rates from four other common forms of cancer, rectal cancer, stomach cancer, lung cancer, and breast cancer, as well as ischemic heart disease. We find much smaller coefficients and no positive, statistically significant relationships in these specifications.

1

Our study is the first to analyze the potential link between brain cancer and mobile phone use on the aggregate, across multiple countries and over a period spanning more than 20 years. This identification strategy allows us to overcome some of the difficulties faced in earlier work. First, we can control for other country-level characteristics that may influence the prevalence of brain cancer and its mortality rate. Second, our aggregate measure of brain cancer may allow us to identify effects that may only be detectable in the aggregate and over a period of multiple decades. Third, an aggregate measure of mobile phone use (mobile phone subscription rates) is not subject to recall problems that may bias case-control studies. The remainder of the paper is organized as follows. Section 2 summaries the previous literature examining the relationship between brain tumors and mobile phone use; Section 3 summarizes our data sources and identification strategy; Section 4 summarizes our results; and Section 5 concludes.

2. Related literature Our paper contributes to an emerging literature on the relationship between mobile phone use and brain cancer. Three types of studies have been undertaken to explore this relationship: (1) brain-imaging studies, (2) case-control studies, and (3) prospective observational studies. In the imaging studies, researchers take scans of the head, brain, and neck of test subjects when a mobile phone that is applied to their ear is turned on or left off, and look for differences in brain metabolism or in radiation absorption. Gandhi et al. (1996) find a high rate of absorption of electromagnetic radiation from a mobile phone of frequency 835 Mhz in the head and neck, and significantly higher absorption in the head and neck of a child than in that of an adult. Using a Samsung Model SCH-U210 mobile phone of frequency 837.5 Mhz, Volkow et al. (2011) find

2

significantly greater brain glucose metabolism rates in the orbital-frontal cortex, particularly on the side of the brain closest to where the mobile phone is applied. However, the clinical significance or implications for the development of brain cancer of these results is unknown. In case-control studies, researchers interview a group of people who have a brain tumor and a group of people who do not have one about their mobile phone use, and they look for differences across the two groups. Coureau et al. (2014) conducted a case-control study in four areas of France from 2004-2006. Using a network of practitioners, they identified 253 people with glioma and 194 people with meningioma (two types of brain tumors that are usually malignant), and they selected 892 matched controls from local electoral roles. They gathered data on mobile phone use through face-to-face interviews. Their findings indicate a statistically significant positive association between heavy mobile phone use (in terms of life-long cumulative duration) and both glioma and meningioma. In case-control studies in Sweden, Hardell et al. (1999; 2013), Hardell et al. (2005), Hardell et al. (2011), and Hardell and Carlberg (2014) also find statistically significant positive associations between mobile phone use and various types of brain tumors. Hardell et al. (2011) perform a pooled analysis of three case-control studies with a total of 1,251 patients diagnosed during 1997-2003 and 2,438 population-based matched controls, while Hardell and Carlberg (2014) perform a pooled analysis of two studies with a total of 1,498 patients diagnosed during 1997–2003 and 2007–2009 and 3,530 matched controls. Both pooled analyses find that mobile phone use is associated with an increased risk of glioma, particularly glioma in the temporal lobe. Interphone Study Group (2010) undertook the largest single case-control study to date, involving 2,708 glioma and 2,409 meningioma cases and matched controls in 13 countries. Overall, they do not find that mobile phone use is associated with an increased risk of glioma or

3

meningioma. However, they find evidence that suggests an increased risk of glioma for people with long-term heavy use of mobile phones. They conclude that “the possible effects of longterm heavy use of mobile phones require further investigation.” In a meta-analysis of 23 case-control studies involving 12,344 cases of malignant and benign brain tumors and 25,572 controls (including the Swedish and Interphone studies), Myung et al. (2009) do not find a statistically significant correlation between mobile phone use and tumors. However, when focusing on the 8 studies that used blinding to the status of patient cases or controls at the interview and on the 10 studies that were judged to have the highest overall methodological quality, they do find a statistically significant positive association. Little et al. (2012) assess the validity of results from case-control studies by comparing them with observed trends in brain cancer incidence and mobile phone use over time. They compare observed and predicted trends in glioma incidence rates in the U.S. over the period 1997-2008, where the predicted rates are calculated using the relative risks of glioma by cumulative mobile phone use estimated by Hardell et al. (2011) and Interphone Study Group (2010). They find that the observed trends in glioma incidence in the U.S. are not consistent with the increased risks of glioma from mobile phone use reported by Hardell et al. (2011). Specifically, based on these reported relative risks, predicted glioma rates are calculated to be 40 percent higher than observed rates in 2008. However, they find that observed trends in glioma could be consistent with the smaller increased risks from long-term heavy use of mobile phones reported in Interphone Study Group (2010). The main problem with most case-control studies is that the results may be subject to recall bias. People who had developed brain cancer may seek to blame their condition on something and therefore may over-remember their mobile phone use when asked about it. This

4

problem is exacerbated in some studies where data on mobile-phone use is collected from relatives after those who had developed brain cancer have passed away. It may be hard to remember one’s own mobile phone use over the past decade, let alone some else’s. In prospective observational studies, researchers select a group of people and collect data on their mobile phone usage, and then follow them through time to see who develops brain cancer. Two large-scale prospective observational studies have been conducted to date: a Danish cohort study (Frei et al. 2011) and a UK study involving 1 million women (Benson et al. 2013). Frei et al. (2011) followed effectively all Danes aged 30 and above and born in Denmark after 1925. They classified the individuals as subscribers or non-subscribers of mobile phones prior to 1995 using administrative data from phone companies, and then determined who among them subsequently developed a brain tumor. No increased risk of developing tumors was found for subscribers, even those who had more than 13 years of recorded subscriptions prior to 1995. One problem with the study is that data on subscriptions were available only until 1995. Individuals with no subscription before 1995 might have had a subscription after 1995 (and may even have had greater usage rates than those who had earlier subscriptions), so the control group is potentially contaminated with users. Benson et al. (2013) surveyed 1 million women in the UK between 1999 and 2005, asking them “how often do you use a mobile phone?” (“never”, “less than once a day,” and “every day”) and “for how long have you used one?” The researchers then determined who developed a brain tumor in a 7-year follow-up. They found no increased risk of glioma or meningioma, but an increased risk of acoustic neuroma (a type of brain cancer that is usually benign) for long-term users as compared with never users.

5

The main difficulty with prospective observational studies of the link between mobile phone use and brain cancer is that brain cancer is extremely rare and may take very long to develop. For example, the chance of developing a glioma every year is about 3 in 100,000 (Hess et al. 2004). Moreover, cancers may take several decades to develop from the time of first exposure to radiation (Coggle and Lindop 1982; Little 2006). The same is true for exposure to some other types of carcinogens. For example, peak national exposure to cigarettes preceded by 35 years the peak in lung cancer mortality (Walker and Brin 1988). To have a reasonable chance of detecting a potential effect of mobile-phone use on brain cancer rates, prospective observational studies would have to follow hundreds of thousands of people for several decades, which may be prohibitively costly. Our study avoids several of the difficulties associated with the existing observational studies as well as case-control studies. Our study is the first to analyze the potential link between brain cancer and mobile phone use on the aggregate, across multiple countries and over a period spanning over two decades. Given how rare brain cancer is and how long it may take to develop, any potential effect of mobile phone use on brain cancer may only be detectable in the aggregate and over a period of multiple decades. Moreover, our study uses an aggregate measure of mobile phone use (mobile phone subscription rates) that is not subject to recall problems that may bias case-control studies.

3. Data and methods We compile data on mortality, mobile phone use, and other country characteristics from a number of sources.

We collect data on cause-specific mortality from the World Health

Organization (WHO) Mortality Database.

This database contains yearly counts of deaths,

6

categorized by ICD codes. We use raw data files to tabulate deaths by a number of causes both possibly affected by radiation from mobile phone use and not likely to be affected by radiation from mobile phone use. Specifically, we examine brain tumors, rectal cancer, stomach cancer, breast cancer, lung cancer and ischemic heart disease.1 We collect information on annual mobile cellular subscriptions per 100 people from the World Bank. The first mobile cellular phones were introduced in the Nordic countries in the early 1980s, and by 1990, Iceland, Norway, and Sweden had more than three mobile phones subscriptions per 100 people. Additionally, the United States, Hong Kong, Singapore, Kuwait, and New Zealand were relatively early adopters, with more than one subscription per 100 people by 1990. By 1995, the European Union had an average of over four subscriptions per 100 people, growing to over 53 subscriptions per 100 people in 2000. North American and high income Asian and Middle Eastern countries followed similar patterns. Besides radiation to the head and genetic risk, other commonly proposed risk factors for brain cancer include cigarette smoking, alcohol consumption, and environmental toxins. Existing studies, however, have found either no evidence or very weak evidence of a link between these other risk factors and brain cancer (Galeone et al. 2013; Calderón-Garcidueñas et al. 2014; Vida et al. 2014). Nonetheless, we collect data on proxies for these other risk factors, since they may affect general mortality, as well as data on several other controls. We collect data on smoking prevalence per capita from the Institute for Health Metrics and Evaluation’s Global Tobacco Trends, alcohol use per capita from the World Health Organization, and carbon dioxide emissions from the World Bank. We also collect health expenditures as a percent of GDP, real

1

During our time period, many countries switch from coding diseases using ICD8, ICD 9 and ICD 10 codes. We use the WHO mortality database codebook to assign deaths to ICD codes. Appendix Table 1 details the codes used for each disease under each ICD coding scheme.

7

GDP per capita, internet use, government spending on education and demographic information from the World Bank. For our main analysis, we run standard panel regression models as follows: log⁡𝑀𝑐𝑡 = 𝛼 + 𝛽𝐶 𝐶𝑒𝑙𝑙𝑐𝑡−𝑙 + 𝑿𝒄𝒕 𝜷𝑿 + 𝜎𝑐 + 𝜏𝑡 + 𝜑𝑐𝑡 + 𝜀𝑐𝑡 ,

(1)

where log𝑀𝑐𝑡 is the logged cause-specific mortality rate for country 𝑐 at year 𝑡; 𝐶𝑒𝑙𝑙𝑐𝑡−𝑙 is the number of mobile phone subscriptions per 100 people for country 𝑐 at year 𝑡 − 𝑙; 𝑿𝒄𝒕 is a vector of contemporaneous controls discussed above—smoking prevalence per capita, alcohol use per capita, carbon dioxide emissions, health expenditures as a percent of GDP, real GDP per capita, internet use, government spending on education, and the percent of the population that is female, between the ages of 15 and 65, and over 65; 𝜎𝑐 and 𝜏𝑡 are country and year fixed effects; 𝜑𝑐𝑡 represents country-specific linear time trends; and 𝜀𝑐𝑡 is an error term.2

We weight our

regressions by country population and cluster our standard errors at the country level to control for autocorrelation (Bertrand et al. 2004). In robustness tests, we also run models with the other commonly proposed risk factors for brain cancer (smoking prevalence per capita, alcohol use per capita, and carbon dioxide emissions) lagged 𝑙 years. We discuss these and other robustness checks in our results section. We use data on cause-specific mortality rates and contemporaneous controls from 1990 to 2012. In this time period, we have data for 84 countries adding up to 1,447 observations, although not all countries have data for all years. Appendix Table 2 shows the countries and years in our sample. For independent variables that have missing data over some years, we linearly impute these missing observations. We drop countries with fewer than five years of mortality data. For cell phone subscriptions, we choose lags at five-year intervals between five 2

Since there are some countries with no cause-specific mortality in a year, we add one to the number of deaths before calculating the rate.

8

and 25 years. Thus, for a 15-year lag, mortality rates in 2012 are linked with cell phone subscriptions from 1997, and mortality rates in 1990 are linked with cell phone subscriptions in 1975. Although data on cause-specific mortality stretches further back in time, there is no variation in cell phone subscriptions before 1980 and several of our control variables that are lagged in some of our specifications are not available before 1965.

Thus, beginning our

examination of mortality in 1990 allows us to keep our sample of countries standard across our different specifications. In robustness tests, we also run models that only include cell phone data after 1980 and models that only include mortality data after the year 2000.

4. Results Table 1 shows summary statistics from all variables in our sample over the period from 1990 to 2012. On average, the annual brain cancer death rate is about 3.7 individuals per 100,000. There are about 50 mobile phone subscriptions per 100 people on average. However there is a lot of temporal variation in mobile phone subscriptions, which we detail more below. To preview our results, we show a simple time series graph of brain cancer mortality and mobile phone subscriptions per 100 people lagged by 15 years in Figure 1. The graph shows that brain cancer deaths reverse a decade long decline around 2000, corresponding to the first increases in mobile phone prevalence beginning in 1985. Table 2 shows the main results from our regression models. Each column shows results lagging mobile phone subscriptions by a certain number of years. Each regression shows coefficients, standard errors in parentheses, and elasticities in brackets. We find no statistically significant relationship between mobile phone subscriptions and brain cancer death rates for 5year, 10-year, and 25-year lags, and we find larger, in absolute value, and statistically significant

9

coefficients for 15-year and 20-year lags. While the coefficient for the 20-year lag is statistically significant only at the 10 percent level, the coefficient on the 15-year lag is significant at the one percent level. Since our dependent variable is logged, our coefficients can be interpreted as semielasticities, so every additional mobile phone subscription per 100 people is associated with a 0.8 percent increase in the brain cancer death rate 15 years later. Transforming this coefficient into a marginal effect by multiplying the semi-elasticity by the mean brain cancer deaths per 100,000 people, we find that one more mobile phone subscription per 100 people is associated with 0.03 more brain cancer deaths per 100,000 people 15 years later. To further put this into context, we can transform the coefficient into an elasticity by multiplying the semi-elasticity by the mean number of mobile phone subscriptions per 100 people. Since each lag period considers mobile phone use over a different time period, the average number of mobile phone subscriptions per 100 people varies. In terms of an elasticity, a one percent increase in mobile phone subscriptions per 100 people is associated with a 0.008 percent increase in the brain cancer death rate 15 years later. We run a number of similar models to test the robustness of our results, including models not logging the dependent variable, not including country-specific time trends, and not using sample weights. Results from these models are presented in Appendix Table 3.

In most

specifications, we find a similar pattern of larger effects and statistical significance for 15 to 25 year lags. We also run various specifications where we include mortality data for 2013 (only available for a subset of countries), lag other possible carcinogens (alcohol use, smoking prevalence, and CO2 emissions) instead of using their cotemporaneous values, only examine countries with at most four than missing observations for mortality, only examine cell phone data after 1980, only examine mortality data after 2000, do not interpolate missing cell phone data, or

10

do not interpolate missing data in any independent variables. In these specifications we see the same pattern of positive and statistically significant associations between cell phone subscriptions rates and brain cancer mortality rates as in our main specification. We also split our sample into current OECD and non-OECD countries and find similar sized marginal effects for both groups of countries. These results are presented in Appendix Table 4. We further test our model by examining the association between lagged mobile phone subscriptions per 100 people and mortality from other common forms of cancer, rectal cancer, stomach cancer, lung cancer, and breast cancer, as well as ischemic heart disease. The mortality rates from these diseases can be seen in the lower panel of Table 1. Table 3 shows results from these regression models. In general, the estimated elasticities are much smaller than the 15 year lag elasticity in Table 2 and often times negative. Unlike in Table 2, no model displays statistical significance at the one percent level, except for breast cancer where the coefficient indicates a negative relationship.

5. Conclusion This paper provides new evidence on the relationship between mobile phone use and brain tumors. We use data on country-level brain cancer death rates from 1990 to 2012 collected by the World Health Organization and mobile phone subscriptions collected by the World Bank. The aggregate, country-level data allows us to potentially uncover effects that may not be detectable in smaller studies.

Moreover, using aggregate mobile phone subscriptions data

circumvents concerns about recall bias present in some previous work. Examining changes in brain cancer death rates and lagged changes in mobile phone subscriptions within countries over time, we find a statistically significant relationship between

11

mobile phone subscriptions and brain cancer death rates 15 years later. Our estimated effect sizes are small but still statistically significant. To err on the side of caution, individuals may want to make greater use of the speaker phone and/or texting options on their cell phones or investigate cell phone cases that may deflect radiation away from their ears. Some caveats to our analysis are warranted. First, we cannot make an unambiguous claim of causality based on our results. Although we control for other characteristics likely to affect brain cancer death rates and find limited statistically significant relationships between mobile phone subscriptions and other mortality, there may be yet other factors that are correlated with mobile phone subscriptions and are also drivers of changes in brain cancer death rates. Second, our measure of mobile phone use, mobile phone subscriptions per 100 people, may not be perfectly accurate. It could be that actual mobile phone use increases as a higher percent of the population adopts mobile phones since there are more people to talk to. However, it could also be that the first adopters of mobile phones are those people who will use mobile phones the most, and thus average mobile phone use may decrease as subscriptions increase. Third, long lags between mobile phone subscriptions and brain cancer death rates mean that, depending on the specified lag between mobile phone subscriptions and mortality, the large recent variations in mobile phone adoption may not be part of our analysis. However, as new data become available, it will be possible to connect this recent variation in mobile phone use with mortality 15 years later. Our study clearly calls for more work in this area.

12

References Abouk, R. and S. Adams (2013). "Texting bans and fatal accidents on roadways: Do they work? Or do drivers just react to announcements of bans?" American Economic Journal: Applied Economics 5(2): 179-199. Asbridge, M., J. R. Brubacher and H. Chan (2013). "Cell phone use and traffic crash risk: a culpability analysis." International Journal of Epidemiology 42(1): 259-267. Beland, L.-P. and R. Murphy (2015). "Ill Communication: Technology, Distraction & Student Performance." Working Paper. Benson, V. S., K. Pirie, J. Schüz, et al. (2013). "Mobile phone use and risk of brain neoplasms and other cancers: prospective study." International Journal of Epidemiology 42(3): 792802. Bertrand, M., E. Duflo and S. Mullainathan (2004). "How Much Should We Trust Differencesin-Differences Estimates?" Quarterly Journal of Economics 119(1): 249-275. Calderón-Garcidueñas, L., A. Calderón-Garcidueñas, R. Torres-Jardón, et al. (2014). "Air pollution and your brain: what do you need to know right now." Primary Health Care Research and Development 26: 1-17. Coggle, J. and P. J. Lindop (1982). Medical consequences of radiation following a global nuclear war. The Aftermath: The Human and Ecological Consequences of Nuclear War. New York, NY, Pantheon Books. Coureau, G., G. Bouvier, P. Lebailly, et al. (2014). "Mobile phone use and brain tumours in the CERENAT case-control study." Occupational and Environmental Medicine 71(7): 514522. Frei, P., A. H. Poulsen, C. Johansen, et al. (2011). "Use of mobile phones and risk of brain tumours: update of Danish cohort study." British Medical Journal 343. Galeone, C., S. Malerba, M. Rota, et al. (2013). "A meta-analysis of alcohol consumption and the risk of brain tumours." Annals of Oncology 24(2): 514-523. Gandhi, O. P., G. Lazzi and C. M. Furse (1996). "Electromagnetic absorption in the human head and neck for mobile telephones at 835 and 1900 MHz." Microwave Theory and Techniques, IEEE Transactions on 44(10): 1884-1897. Hardell, L. and M. Carlberg (2014). "Mobile phone and cordless phone use and the risk for glioma–Analysis of pooled case-control studies in Sweden, 1997–2003 and 2007–2009." Pathophysiology 22(1): 1-13. Hardell, L., M. Carlberg and K. Hansson Mild (2005). "Case-control study on cellular and cordless telephones and the risk for acoustic neuroma or meningioma in patients diagnosed 2000-2003." Neuroepidemiology 25(3): 120-128. Hardell, L., M. Carlberg and K. Hansson Mild (2011). "Pooled analysis of case-control studies on malignant brain tumours and the use of mobile and cordless phones including living and deceased subjects." International Journal of Oncology 38(5): 1465-1474. Hardell, L., M. Carlberg, F. Söderqvist, et al. (2013). "Pooled analysis of case-control studies on acoustic neuroma diagnosed 1997-2003 and 2007-2009 and use of mobile and cordless phones." International Journal of Oncology 43(4): 1036-1044. Hardell, L., A. Näsman, A. Påhlson, et al. (1999). "Use of cellular telephones and the risk for brain tumours: A case-control study." International Journal of Oncology 15(1): 113-116.

13

Hess, K. R., K. R. Broglio and M. L. Bondy (2004). "Adult glioma incidence trends in the United States, 1977–2000." Cancer 101(10): 2293-2299. Interphone Study Group (2010). "Brain tumour risk in relation to mobile telephone use: results of the INTERPHONE international case–control study." International Journal of Epidemiology 39(3): 675-694. Kolko, J. D. (2009). "The effects of mobile phones and hands-free laws on traffic fatalities." The BE Journal of Economic Analysis & Policy 9(1). Kuznekoff, J. H. and S. Titsworth (2013). "The impact of mobile phone usage on student learning." Communication Education 62(3): 233-252. Little, J. B. (2006). Chapter 19: Ionizing Radiation, In Kufe et al., Cancer Medicine (6th ed.). Hamilton, Ontario, B.C. Decker. Little, M., P. Rajaraman, R. Curtis, et al. (2012). "Mobile phone use and glioma risk: comparison of epidemiological study results with incidence trends in the United States." British Medical Journal 344. Myung, S.-K., W. Ju, D. D. McDonnell, et al. (2009). "Mobile phone use and risk of tumors: a meta-analysis." Journal of Clinical Oncology 27(33): 5565-5572. Thornton, B., A. Faires, M. Robbins, et al. (2015). "The Mere Presence of a Cell Phone May be Distracting." Social Psychology. Vida, S., L. Richardson, E. Cardis, et al. (2014). "Brain tumours and cigarette smoking: analysis of the INTERPHONE Canada case–control study." Environmental Health 13(1): 55. Volkow, N. D., D. Tomasi, G.-J. Wang, et al. (2011). "Effects of cell phone radiofrequency signal exposure on brain glucose metabolism." Journal of the American Medical Association 305(8): 808-813. Walker, W. J. and B. N. Brin (1988). "US lung cancer mortality and declining cigarette tobacco consumption." Journal of Clinical Epidemiology 41(2): 179-185.

14

3.2

3.4

3.6

3.8

4

1 2 3 Lagged Mobile Phone Subscriptions per 100

Brain Cancer Mortaility Rate per 100,000 People Lagged Mobile Phone Subscriptions per 100 People

0

3

Brain Cancer Mortaility Rate per 100,000

4.2

4

Figure 1: Time Series of Global Brain Cancer Motaility Rate per 100,000 People and Global Mobile Phone Subscriptions per 100 People Lagged by 15 Years

1990

1995

2000 Year

Notes: Data from the WHO Mortality Database and the World Bank.

15

2005

2010

Table 1 Summary Statistics (N=1447) Variable Brain Cancer Mortality Rate Cell Phone Prevalence Smoking Prevalence

Description Brain Cancer Mortality Rate per 100,000 Mobile cellular subscriptions (per 100 people) Smoking prevalence for individuals ages 15 and older

Internet Use Health Spending/Capita Real GDP/Capita Education Spending/GDP

Internet users (per 100 people) Health expenditure per capita (current US$) GDP per capita (current US$) Total government spending on public and private education as a percent of GDP Recorded alcohol per capita (15+ years) consumption of pure alcohol (liters) CO2 emissions (metric tons per capita) Percent of a country's population that is female Percent of a country's population that is over age 65 Percent of a country's population that is between age 15 and 65 Rectal Cancer Mortality Rate per 100,000 Stomach Cancer Mortality Rate per 100,000 Lung Cancer Mortality Rate per 100,000 Breast Cancer Mortality Rate per 100,000 Isachemic Heart Disease Mortality per 100,000

Alcohol Prevalence CO2 Emissions Percent Female Percent Over 65 Percent 15 to 65 Rectal Cancer Mortality Rate Stomach Cancer Mortality Rate Lung Cancer Mortality Rate Breast Cancer Mortality Rate Heart Disease Mortality

Notes: Sample statistics are weighted by country population.

16

Source World Health Organization World Bank Global Health Data Exchange and Institute for Health Metrics and Evaluation World Bank World Bank World Bank World Bank

Mean 3.66 50.11 21.92

Std. Dev. 1.96 42.68 6.64

28.97 2113.39 20040.41 4.69

27.86 2107.80 15544.61 1.10

7.98

3.04

World Bank World Bank World Bank

8.58 50.87 11.40

6.11 0.88 5.11

World Bank

65.89

3.10

17.69 11.59 33.95 22.32 104.44

11.72 10.30 21.17 13.33 76.35

World Health Organization

World Health Organization World Health Organization World Health Organization World Health Organization World Health Organization

Table 2 Lagged Cell Phone Prevalence and Brain Cancer Mortality 1990 to 2012

5 Years

10 Years

Cell Phone Prevalence

0.000 (0.001) [0.000]

0.002 (0.001) [0.017]

N Adjusted R-Squared Dep Var Mean Cell Phone Mean

1447 0.982 1.106 24.699

1447 0.982 1.106 8.728

Lags 15 Years 0.008 *** (0.002) [0.011] 1447 0.982 1.106 1.369

20 Years 0.044 * (0.025) [0.008] 1447 0.982 1.106 0.172

25 Years -0.007 (0.008) [0.000] 1447 0.981 1.106 0.023

Notes: Data from sources described in Table 1. The table shows coefficients, standard errors clustered at the country level in parentheses, and elasticities in brackets. In addition to the coefficients shown, all models include smoking prevalence, alcohol use, internet use, CO2 emissions, health spending per capita, GDP per capita, education spending/GDP, the percent of the population that is female, between the ages of 15 and 65, and over 65, country and year fixed effects, and country specific linear time trends. Regressions are weighted by country population. Stars denote statistical significance: * Significant at 10% level, ** Significant at 5% level, and *** Significant at 1% level.

17

Table 3 Lagged Cell Phone Prevalence and Other Cause Mortality 1990 to 2012

5 Years

10 Years

Lags 15 Years

20 Years

25 Years

Rectal Cancer

0.001 (0.001) [0.025]

0.000 (0.001) [0.000]

0.001 (0.001) [0.001]

0.031 (0.021) [0.005]

-0.002 (0.011) [0.000]

Stomach Cancer

0.000 (0.001) [0.000]

0.001 (0.001) [0.009]

-0.001 (0.002) [-0.001]

0.017 (0.018) [0.003]

0.003 (0.008) [0.000]

Lung Cancer

0.000 (0.000) [0.000]

0.001 (0.001) [0.009]

0.000 (0.001) [0.000]

0.026 (0.017) [0.004]

0.003 (0.008) [0.000]

Breast Cancer

0.000 (0.000) [0.000]

0.000 (0.000) [0.000]

-0.004 (0.012) [-0.001]

-0.004 (0.005) [0.000]

Heart Disease

-0.002 (0.001) [-0.049]

-0.001 (0.002) [-0.009]

-0.001 (0.002) [-0.001]

0.001 (0.027) [0.000]

-0.024 * (0.014) [-0.001]

24.699

8.728

1.369

0.172

Cell Phone Mean

-0.002 ** (0.001) [-0.003]

0.023

Notes: Data from sources described in Table 1. The table shows coefficients, standard errors clustered at the country level in parentheses, and elasticities in brackets. In addition to the coefficients shown, all models include smoking prevalence, alcohol use, internet use, CO2 emissions, health spending per capita, GDP per capita, education spending/GDP, the percent of the population that is female, between the ages of 15 and 65, and over 65, country and year fixed effects, and country specific linear time trends. Regressions are weighted by country population. Stars denote statistical significance: * Significant at 10% level, ** Significant at 5% level, and *** Significant at 1% level.

18

Appendix Tables

19

Appendix Table 1 ICD Codes Used for ICD 8, ICD 9, and ICD 10 Brain Cancer

Rectal Cancer

Stomach Cancer

Lung Cancer

Breast Cancer

Heart Disease

ICD 8A

191

A049, 154

A047, 151

A051, 162

A054, 174

A083

ICD 9

B130, 191

B093, B094, B091, 151 153, 154

B101, 162

B113, 174

B27, 410, 411, 412, 413, 414

ICD 10

C71

C18, C19, C20, C21

C33, C34

C50

I20, I21, I22, I23, I24, I25

C16

Notes: Data from the World Health Organization.

20

Appendix Table 2 Countries and Years Included in Sample

Country Albania Antigua and Barbuda Argentina Australia Austria Bahrain Barbados Belgium Belize Brazil Bulgaria Canada Chile Colombia Costa Rica Croatia Cuba Cyprus Czech Republic Denmark Dominican Republic Ecuador Egypt El Salvador Estonia Fiji Finland France Georgia Germany Greece Guatemala Guyana Honduras Hungary Iceland Ireland Israel Italy Jamaica Japan Kiribati Kuwait Kyrgyzstan Latvia Lithuania Luxembourg

country 4005 2010 2020 5020 4010 3020 2040 4020 2045 2070 4030 2090 2120 2130 2140 4038 2150 3080 4045 4050 2170 2180 1125 2190 4055 5070 4070 4080 4084 4085 4140 2250 2260 2280 4150 4160 4170 3150 4180 2290 3160 5105 3190 4184 4186 4188 4190

Year 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 X

X

X

X

X

X X

X X

X X

X X

X X

X X

X X

X

X

X

X

X

X

X

X X

X X

X X

X X

X X

X X

X X X

X

X

X

X

X

X

X

X X

X X

X X

X X

X X

X X

X X X

X

X

X X X X X

X X X X X X

X X X X X X

X X X X X X X X X

X X X X X X X X X

X X X X X X X X X

X X X X

X X X X

X X X X

X

X

X

X X

X X

X X X X X X X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X

X X X X X

X

X

X

X

X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X

X X

X X

X X

X X X

X X X

X X X

X X X

X X

X X

X X X

X X X

X X X

X X X

X X X X X X X

X

X

X

X X

X X

21

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X

X X X X X X X X X

X X X X X X X X X X

X X X X

X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X

X X

X X X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X

X X

X X

X X X X X X X X X X X X X X X X X X X X X X X

X X X X X

X X

X X X X X X X

X X X X X X X X X X

X

X

X

X

X

X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X X X

X X X X X X X

X X X X X X X

X X X X X X X

X X X X

X X X X

X X

X X X X X

X X X X X

X X X X X

X X X X X

X X X X X

X X

Total Obs. 18 10 16 21 23 16 12 23 16 17 23 22 16 15 16 23 13 9 23 23 14 16 15 16 19 12 23 22 12 23 23 13 16 6 23 20 21 23 21 7 23 8 18 13 17 20 23

Appendix Table 2 Countries and Years Included in Sample

Country Malaysia Malta Mauritius Mexico Morocco Netherlands New Zealand Nicaragua Norway Panama Paraguay Peru Philippines Poland Portugal Qatar Republic of Korea Republic of Moldova Romania Saint Lucia Saint Vincent and Grenadines Serbia Singapore Slovakia Slovenia South Africa Spain Sweden Switzerland TFYR Macedonia Thailand Trinidad and Tobago United Kingdom United States of America Uruguay Venezuela

country 3236 4200 1300 2310 1310 4210 5150 2340 4220 2350 2360 2370 3300 4230 4240 3320 3325 4260 4270 2400 2420 4273 3350 4274 4276 1430 4280 4290 4300 4195 3380 2440 4308 2450 2460 2470

Year 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

X X

X X

X X

X X

X X

X X

X X

X X

X X X

X X X

X X X X

X X X X

X X X X

X X X X

X X X X

X X X X

X X X X

X X X X

X X

X X

X X

X X

X X

X X

X X

X

X

X

X

X

X

X

X X X X

X

X

X X X X X X

X

X

X

X

X X X X X X X X X X

X X X X X X X

X

X X X X X X X X X X

X X X X X X X

X

X X X X X X X X X X

X X X X X X X

X

X X X X X X X X X X

X X X X X X X

X X X

X X X X X X X X X X

X X X X

X X X X

X X X X

X X X X

X X X

X X X X X X X X X X X X

X X X

X X X X X X X X X X X X X

X X X X X

X X X X X X X

X X X X X X X X X X X X

X X X X X X X

X X X X X X X X X X X X X

X X X X X X

X X X X X X X

X X X X X X X

X X X X X X X

X X X X X X X

X X

X X

X X X

X X X

X X X

X

X X X

X X X

X X X

X X X

X

X X X X X X X

X X

X

X

X

X X X

X X X X

X X X X X

X X X X X

X X

X X

X X

X X X X X X X X

X X X X X X X X X

X X X X X X

X X X X X X

X X X X X X

X X X X X X X

X X

X X

X X

X X

X X

X X

X X

X X

X X X X

X X X X

X

22

X X X X X X

X X X X X X X X X X X X X

X X X X X X X

X X X X X X X X X X X X X X

X X X X X X X X X X X

X X X X X X X X X X X

X X X X X X X X X X X

X X X X X

X X X X X X X X X X X X X X X

X X X X X X X

X X X X X X X X X X

X X X X X X X

X X X X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X

X X X

X

X

X

X

X X X X X

X X X X X

X X X X X

X X X

X

X

X X X X X X X

X X X X X

X X

Total Obs. 9 23 23 15 5 23 22 16 23 15 17 14 13 21 20 10 23 22 23 14 9 15 23 19 21 20 23 23 5 20 15 10 22 21 14 14

Appendix Table 3 Lagged Cell Phone Prevalence and Brain Cancer Mortality - Other Specifications

5 Years

10 Years

Lags 15 Years

20 Years

25 Years

Not Logged

-0.001 (0.003) [-0.007]

0.002 (0.004) [0.005]

0.020 *** (0.007) [0.007]

0.105 (0.088) [0.005]

Unweighted

0.000 (0.001) [0.000]

0.002 (0.002) [0.017]

0.009 * (0.005) [0.012]

0.056 * (0.031) [0.010]

0.002 (0.011) [0.000]

-0.001 (0.001) [-0.025]

0.001 (0.001) [0.009]

0.006 ** (0.003) [0.008]

0.009 (0.009) [0.002]

0.003 (0.007) [0.000]

Not Logged + Unweighted + Not Including Trends

0.002 (0.003) [0.014]

0.004 (0.004) [0.010]

0.013 (0.015) [0.005]

0.091 *** (0.023) [0.004]

0.131 *** (0.020) [0.001]

Dep Var Mean Cell Phone Mean

3.657 24.699

3.657 8.728

3.657 1.369

Not Including Trends

3.657 0.172

-0.086 ** (0.041) [-0.001]

3.657 0.023

Notes: Data from sources described in Table 1. The table shows coefficients, standard errors clustered at the country level in parentheses, and elasticities in brackets. In addition to the coefficients shown, all models include smoking prevalence, alcohol use, internet use, CO2 emissions, health spending per capita, GDP per capita, education spending/GDP, the percent of the population that is female, between the ages of 15 and 65, and over 65, and country and year fixed effects. Stars denote statistical significance: * Significant at 10% level, ** Significant at 5% level, and *** Significant at 1% level.

23

Appendix Table 4 Lagged Cell Phone Prevalence and Brain Cancer Mortality - Other Robustness Checks

5 Years

10 Years

Lags 15 Years

20 Years

25 Years

Add 2013 Data (N=1473)

0.000 (0.001) {25.983} [0.000]

0.001 (0.001) {9.627} [0.010]

0.008 *** 0.036 * (0.002) (0.021) {1.673} {0.201} [0.013] [0.007]

-0.004 (0.010) {0.030} [0.000]

Lag alcohol, tobacco and CO2 (N=1447)

0.000 (0.001) {24.699} [0.000]

0.002 (0.001) {8.728} [0.017]

0.009 *** 0.051 * (0.002) (0.026) {1.369} {0.172} [0.012] [0.009]

-0.009 (0.010) {0.023} [0.000]

Including Countries With At Least 20 Years of Data (N=842)

-0.001 (0.001) {29.488} [-0.029]

-0.001 (0.001) {11.436} [-0.011]

0.007 *** 0.027 (0.002) (0.032) {1.856} {0.206} [0.013] [0.006]

-0.089 (0.071) {0.017} [-0.002]

Only Including Cell Phone Data After 1980 (N Varies)

0.000 (0.001) {21.912} [0.000]

0.002 (0.001) {8.728} [0.017]

0.007 *** 0.028 (0.001) (0.022) {1.620} {0.269} [0.011] [0.008]

-0.095 (0.114) {0.059} [-0.006]

Only Including Mortality Data After 2000 (N=943)

-0.001 ** 0.001 (0.001) (0.001) {37.981} {13.575} [-0.038] [0.014]

0.006 *** 0.028 (0.001) (0.022) {2.134} {0.269} [0.013] [0.008]

-0.006 (0.011) {0.035} [0.000]

No Interpolation of Cell Phone Data (N Varies)

0.000 (0.001) {24.756} [0.000]

0.002 (0.001) {8.937} [0.018]

0.008 *** 0.053 * (0.002) (0.027) {1.355} {0.144} [0.011] [0.008]

-0.015 (0.060) {0.011} [0.000]

No Interpolation of Any Independent Variables (N Varies)

0.000 (0.001) {26.926} [0.000]

0.001 (0.001) {8.036} [0.008]

0.013 *** 0.054 ** (0.005) (0.026) {1.139} {0.149} [0.015] [0.008]

0.012 (0.045) {0.007} [0.000]

OECD Countries (N=696)

-0.001 (0.001) {29.475} [-0.029]

0.000 (0.001) {11.473} [0.000]

0.009 *** 0.044 (0.002) (0.030) {1.846} {0.207} [0.017] [0.009]

-0.026 (0.067) {0.017} [0.000]

Non-OECD Countries (N=751)

0.004 * (0.002) {15.521} [0.062]

0.039 ** (0.017) {0.452} [0.018]

-0.003 (0.015) {0.034} [0.000]

0.006 (0.004) {3.455} [0.021]

0.024 (0.044) {0.105} [0.003]

Notes: Data from sources described in Table 1. The table shows coefficients, standard errors clustered at the country level in parentheses, the mean mobile phone subscriptions in braces, and elasticities in brackets. In addition to the coefficients shown, all models include smoking prevalence, alcohol use, internet use, CO2 emissions, health spending per capita, GDP per capita, education spending/GDP, the percent of the population that is female, between the ages of 15 and 65, and over 65, country and year fixed effects, and country specific linear time trends. Regressions are weighted by country population. Stars denote statistical significance: * Significant at 10% level, ** Significant at 5% level, and *** Significant at 1% level.

24