Small declines in agility, facial features may predict risk of dying (Epidemiology)

Photo source: Shutter Stock

Photo source: Shutter Stock

By B. Rose Huber, Woodrow Wilson School of Public and International Affairs

A new study from Princeton University shows that health assessments made by medically untrained interviewers may predict the mortality of individuals better than those made by physicians or the individuals themselves.

Features like forehead wrinkles and lack of agility may reflect a person’s overall health and risk of dying, according to recent health research. But do physicians consider such details when assessing patients’ overall health and functioning?

In a survey of approximately 1,200 Taiwanese participants, Princeton University researchers found that interviewers — who were not health professionals but were trained to administer the survey — provided health assessments that were related to a survey participant’s risk of dying, in part because they were attuned to facial expressions, responsiveness and overall agility.

The researchers report in the journal Epidemiology that these assessments were even more accurate predictors of dying than assessments made by physicians or even the individuals themselves. The findings show that survey interviewers, who typically spend a fair amount of time observing participants, can glean important information regarding participants’ health through thorough observations.

“Your face and body reveal a lot about your life. We speculate that a lot of information about a person’s health is reflected in their face, movements, speech and functioning, as well as in the information explicitly collected during interviews,” said Noreen Goldman, Hughes-Rogers Professor of Demography and Public Affairs in the Woodrow Wilson School.

Together with lead author of the paper, Princeton Ph.D. candidate Megan Todd, Goldman analyzed data collected by the Social Environment and Biomarkers of Aging Study (SEBAS). This study was designed by Goldman and co-investigator Maxine Weinstein at Georgetown University to evaluate the linkages among the social environment, stress and health. Beginning in 2000, SEBAS conducted extensive home interviews, collected biological specimens and administered medical examinations with middle-aged and older adults in Taiwan. Goldman and Todd used the 2006 wave of this study, which included both interviewer and physician assessments, for their analysis. They also included death registration data through 2011 to ascertain the survival status of those interviewed.

The survey used in the study included detailed questions regarding participants’ health conditions and social environment. Participants’ physical functioning was evaluated through tasks that determined, for example, their walking speed and grip strength. Health assessments were elicited from participants, interviewers and physicians on identical five-point scales by asking “Regarding your/the respondent’s current state of health, do you feel it is excellent (5), good (4), average (3), not so good (2) or poor (1)?”

Participants answered this question near the beginning of the interview, before other health questions were asked. Interviewers assessed the participants’ health at the end of the survey, after administering the questionnaire and evaluating participants’ performance on a set of tasks, such as walking a short distance and getting up and down from a chair. And physicians — who were hired by the study and were not the participants’ primary care physicians — provided their assessments after physical exams and reviews of the participants’ medical histories. (Study investigators did not provide special guidance about how to rate overall health to any group.)

In order to understand the many variables that go into predicting mortality, Goldman and Todd factored into their statistical models such socio-demographic variables as gender, place of residence, education, marital status, and participation in social activities. They also considered chronic conditions, psychological wellbeing (such as depressive symptoms) and physical functioning to account for a fuller picture of health.

“Mortality is easy to measure because we have death records indicating when a person has died,” Goldman said. “Overall health, on the other hand, is very complicated to measure but obviously very important for addressing health policy issues.”

Two unexpected results emerged from Goldman and Todd’s analysis. The first: physicians’ ratings proved to be weak predictors of survival. “The physicians performed a medical exam equivalent to an annual physical exam, plus an abdominal ultrasound; they have specialized knowledge regarding health conditions,” Goldman explained. “Given access to such information, we anticipated stronger, more accurate predictions of death,” she said. “These results call into question previous studies’ assumptions that physicians’ ‘objective health’ ratings are superior to ‘subjective’ ratings provided by the survey participants themselves.”

In a second surprising finding, the team found that interviewers’ ratings were considerably more powerful for predicting mortality than self-ratings. This is likely, Goldman said, because interviewers considered respondents’ movements, appearance and responsiveness in addition to the detailed health information gathered during the interviews. Also, Goldman posits, interviewer ratings are probably less affected by bias than self-reports.

“The ‘self-rated health’ question is religiously used by health researchers and social scientists, and, although it has been shown to predict mortality, it suffers from many biases. People use it because it’s easy and simple,” Goldman continued. “But the problem with self-rated health is that we have no idea what reference group the respondent is using when evaluating his or her own health. Different ethnic and racial groups respond differently as do varying socioeconomic groups. We need other simple ways to rate individual health instead of relying so heavily on self-rated health.”

One way, Goldman suggests, is by including interviewer ratings in surveys along with self-ratings: “This is a straightforward and cost-free addition to a questionnaire that is likely to improve our measurement of health in any population,” Goldman said.

The paper, “Do Interviewer and Physician Health Ratings Predict Mortality? A Comparison with Self-Rated Health,” first appeared online in Epidemology in August 2013. The article also will be featured in the November print edition. The research was conducted with the assistance of colleagues at Princeton’s Office of Population Research, Georgetown University and the Bureau of Health Promotion in the Taiwan Department of Health.

Read the abstract.

Todd MA, Goldman N. Do interviewer and physician health ratings predict mortality?: a comparison with self-rated health. Epidemiology. 2013 Nov;24(6):913-20. doi: 10.1097/EDE.0b013e3182a713a8.

 

Contaminated water linked to low-weight babies, prematurity (Canadian Journal of Economics)

By B. Rose Huber, Woodrow Wilson School of Public and International Affairs

Image: Shutter Stock

Image: Shutterstock.

Pregnant women living in areas with contaminated drinking water may be more likely to have babies that are premature or have low birth weights (less than 5.5 pounds), according to a study based at Princeton University’s Woodrow Wilson School of Public and International Affairs.

Published in the Canadian Journal of Economics, the study shows that the effects of contaminated water—such as cognitive and developmental impairments—are particularly significant for babies born to less-educated mothers. These mothers also are less likely to uproot from areas with contaminated water, which, the authors note, suggests a need for serious improvement in terms of communicating with people living in such environs.

“Fetuses are vulnerable to all types of pollution, including water contamination caused by chemicals and bacteria,” said Janet Currie, the Henry Putnam Professor of Economics and Public Affairs at the Wilson School and director of the Center for Health and Wellbeing. “This contamination can lead to a host of problems, including low-birth-weight babies who can have lifelong cognitive struggles. It’s a particular problem for less-educated women who also presumably have fewer options in terms of housing.”

While other studies have focused on the effects of air pollution on infant health, Currie’s is one of the first to evaluate the effects of water pollution on infants. Together with researchers from Columbia University and the University of California-San Diego, Currie examined ten years of New Jersey birth records and data on drinking-water quality collected from 1997 to 2007. All birth records contained information regarding the date of birth, the infant’s health at birth, and maternal characteristics such as race, education and marital status. To determine whether mothers relocated due to water contamination, the researchers studied sets of siblings and whether mothers moved between births.

Using data from the New Jersey Department of Environmental Protection (DEP), Currie and her team looked at violation records across 488 water districts in New Jersey and found that more than a quarter of districts had water contamination violations affecting more than 30,000 people. These violations included both chemical and bacterial contamination caused by such contaminants as dichloroethane — a solvent often used for plastics or as degreasers — as well as radon and coliform.

The researchers matched the birth records to the water systems that serve the infants’ residences. Because weather can dictate the amount of water a person consumes, they also incorporated daily temperatures into their data set.

“We found that infants exposed to contamination in utero tend to have mothers who are younger, less educated and less likely to be married than other mothers. They are also more likely to be African-American or Hispanic,” Currie said. “The results also suggest that mothers who are less educated are less likely than other mothers to move in response to contamination, while older mothers are more likely to drink bottled water or move.”

Currie notes that when a water district is affected, the DEP is required to send a notice to all residences. However, for renters, there may be routing difficulties.

“If someone puts something in your mailbox, do you even see it? Does your landlord pick it up?” said Currie. “Notices are being sent that people don’t receive. There’s an undercurrent here that the way information is sent isn’t adequate. We need to get this information to people directly.”

Currie suggests that health-care workers include literature about water contamination risks and hazards in clinics and exam rooms to reach more pregnant women.

“If it’s going to be harmful for some groups, we need to at least let those groups know about them, so they can avoid it,” said Currie.

In the future, Currie plans to continue studying environmental impacts on child health while also pursuing the relationship between home foreclosures and health.

Other collaborators for the study include: from Columbia University, Katherine Meckel, Matthew Neidell, and Wolfram Schlenker; and from the University of California, San Diego, Joshua Graff Zivin.

Read the abstract.

Currie, Janet, Joshua Graff Zivin, Katherine Meckel, Matthew Neidell, and Wolfram Schlenker. August 2013.  Something in the water: contaminated drinking water and infant health. Canadian Journal of Economics. Vol. 46, No. 3, pages 791-810.

Funding was provided by the John D. and Catherine T. MacArthur Foundation, the Environmental Protection Agency and the National Science Foundation.

Nano-dissection identifies genes involved in kidney disease (Genome Research)

Scanning electron microscope (SEM) micrograph of podocytes

Researchers at Princeton and the University of Michigan have created a computer-based method for separating and identifying genes from diseased kidney cells known as podocytes, pictured above. (Image courtesy of Matthias Kretzler)

By Catherine Zandonella, Office of the Dean for Research

Understanding how genes act in specific tissues is critical to our ability to combat many human diseases, from heart disease to kidney failure to cancer.  Yet isolating individual cell types for study is impossible for most human tissues.

A new method developed by researchers at Princeton University and the University of Michigan called “in silico nano-dissection” uses computers rather than scalpels to separate and identify genes from specific cell types, enabling the systematic study of genes involved in diseases.

The team used the new method to successfully identify genes expressed in cells known as podocytes — the “work-horses” of the kidney — that malfunction in kidney disease. The investigators showed that certain patterns of activity of these genes were correlated with the severity of kidney impairment in patients, and that the computer-based approach was significantly more accurate than existing experimental methods in mice at identifying cell-lineage-specific genes. The study was published in the journal Genome Research.

Using this technique, researchers can now examine the genes from a section of whole tissue, such as a biopsied section of the kidney, for specific signatures associated with certain cell types. By evaluating patterns of gene expression under different conditions in these cells, a computer can use machine-learning techniques to deduce which types of cells are present. The system can then identify which genes are expressed in the cell type in which they are interested.  This information is critical both in defining novel disease biomarkers and in selecting potential new drug targets.

By applying the new method to kidney biopsy samples, the researchers identified at least 136 genes as expressed specifically in podocytes. Two of these genes were experimentally shown to be able to cause kidney disease. The authors also demonstrated that in silico nano-dissection can be used for cells other than those found in the kidney, suggesting that the method is useful for the study of a range of diseases.

The computational method was significantly more accurate than another commonly used technique that involves isolating specific cell types in mice. The nano-dissection method’s accuracy was 65% versus 23% for the mouse method, as evaluated by a time-consuming process known as immunohistochemistry which involves staining each gene of interest to study its expression pattern.

The research was co-led by Olga Troyanskaya, a professor of computer science and the Lewis-Sigler Institute for Integrative Genomics at Princeton, and Matthias Kretzler, a professor of computational medicine and biology at the University of Michigan. The first authors on the study were Wenjun Ju, a research assistant professor at the University of Michigan, and Casey Greene, now at the Geisel School of Medicine at Dartmouth and a former postdoctoral fellow at Princeton.

The research was supported in part by National Institutes of Health (NIH) R01 grant GM071966 to OGT and MK, by NIH grants RO1 HG005998 and DBI0546275 to OGT, by NIH center grant P50 GM071508, and by NIH R01 grant DK079912 and P30 DK081943 to MK. OGT also receives support from the Canadian Institute for Advanced Research.

Read the abstract.

Wenjun Ju, Casey S Greene, Felix Eichinger, Viji Nair, Jeffery B Hodgin, Markus Bitzer, Young-suk Lee, Qian Zhu, Masami Kehata, Min Li, Song Jiang, Maria Pia Rastaldi, Clemens D Cohen, Olga G Troyanskaya and Matthias Kretzler. 2013. Defining cell-type specificity at the transcriptional level in human disease. Genome Research. Published in Advance August 15, 2013, doi: 10.1101/gr.155697.113.

New mouse model for hepatitis C (Nature)

By Catherine Zandonella, Office of the Dean for Research

Hepatitis C affects about three million people in the U.S. and is a leading cause of chronic liver disease, so creating a vaccine and new treatments is an important public health goal. Most research to date has been done in chimpanzees because they are one of a handful of species that become infected and spread the virus.

Now researchers led by Alexander Ploss of Princeton University and Charles Rice of the Rockefeller University have generated a mouse that can become infected with hepatitis C virus (HCV).  They reported the advance in the Sept 12 issue of the journal Nature. “The entire life cycle of the virus — from infection of liver cells to viral replication, assembly of new particles, and release from the infected cell — occurs in these mice,” said Ploss, who joined the Princeton faculty in July 2013 as assistant professor of molecular biology.

Ploss and his colleagues have been working for some time on the challenge of creating a small animal model for studying the disease. Four years ago, while at the Rockefeller University in New York, Ploss and Rice identified two human proteins, known as CD81 and occludin, that enable mouse cells to become infected with HCV (Nature 2009). In a follow up study Ploss and colleagues showed that a mouse engineered to express these human proteins could become infected with HCV, although the animals could not spread the virus (Nature 2011).

In the present study, which included colleagues at Osaka University and the Scripps Research Institute, the researchers bred the human-protein-containing mice with another strain that had a defective immune system – one that could not easily rid the body of viruses. The resulting mice not only become infected, but could potentially pass the virus to other susceptible mice.

The availability of this new way to study HCV could help researchers discover new vaccines and treatments, although Ploss cautioned that more work needs to be done to refine the model.

The study was supported in part by award number RC1DK087193 from the National Institute of Diabetes and Digestive and Kidney Diseases; R01AI072613, R01AI099284, and R01AI079031 from the National Institute for Allergy and Infectious Disease; R01CA057973 from the National Cancer Institute; and several foundations and contributors, as well as the Infectious Disease Society of America and the American Liver Foundation.

Read the abstract

Marcus Dorner, Joshua A. Horwitz, Bridget M. Donovan, Rachael N. Labitt, William C. Budell, Tamar Friling, Alexander Vogt, Maria Teresa Catanese, Takashi Satoh, Taro Kawai, Shizuo Akira, Mansun Law, Charles Rice & Alexander Ploss. 2013. Completion of the entire hepatitis C virus life cycle in genetically humanized mice. Nature 501, 237–241 (First published online on 31 July 2013)  doi:10.1038/nature12427.

 

Shingles symptoms may be caused by neuronal short circuit (Proceedings of the National Academy of Sciences)

By Catherine Zandonella, Office of the Dean for Research

Neurons firing in synchrony could be responsible for pain, itch in shingles and herpes infection. Click to view movie. (Source: PNAS)

The pain and itching associated with shingles and herpes may be due to the virus causing a “short circuit” in the nerve cells that reach the skin, Princeton researchers have found.

This short circuit appears to cause repetitive, synchronized firing of nerve cells, the researchers reported in the journal Proceedings of the National Academy of Sciences. This cyclical firing may be the cause of the persistent itching and pain that are symptoms of oral and genital herpes as well as shingles and chicken pox, according to the researchers.

These diseases are all caused by viruses of the herpes family. Understanding how these viruses cause discomfort could lead to better strategies for treating symptoms.

The team studied what happens when a herpes virus infects neurons. For research purposes the investigators used a member of the herpes family called pseudorabies virus. Previous research indicated that these viruses can drill tiny holes in neurons, which pass messages in the form of electrical signals along long conduits known as axons.

The researchers’ findings indicate that electrical current can leak through these holes, or fusion pores, and spread to nearby neurons that were similarly damaged, causing the neurons to fire all at once rather than as needed. The pores were likely created for the purpose of infecting new cells, the researchers said.

Researchers at Princeton University imaged the synchronized, repetitive firing of herpes-infected neurons in a region known as the submandibular ganglia (SMG) between the salivary glands and the brain in mice. Image source: PNAS.

Researchers at Princeton University imaged the synchronized, repetitive firing of herpes-infected neurons in a region known as the submandibular ganglia (SMG) between the salivary glands and the brain in mice. (Source: PNAS)

The investigators observed the cyclical firing of neurons in a region called the submandibular ganglia between the salivary glands and the brain in mice using a technique called 2-photon microscopy and dyes that flash brightly when neurons fire. (Movie of synchronized firing of herpes-infected neurons.)

The team found that two viral proteins appear to work together to cause the simultaneous firing, according to Andréa Granstedt, who received her Ph.D. in molecular biology at Princeton in 2013 and is the first author on the article.  The team was led by Lynn Enquist, Princeton’s Henry L. Hillman Professor in Molecular Biology and a member of the Princeton Neuroscience Institute.

Each colored line and number on the right represents an individual neuron. The overlapping peaks indicate synchronized firing of neurons, which occurs when electrical current is able to leak from one neuron to the next. (Source: PNAS)

The first of these two proteins is called glycoprotein B, a fusion protein that drills the holes in the axon wall. A second protein, called Us9, acts as a shuttle that sends glycoprotein B into axons, according to the researchers. “The localization of glycoprotein B is crucial,” Granstedt said. “If glycoprotein B is present but not in the axons, the synchronized flashing won’t happen.”

The researchers succeeded in stopping the short circuit from occurring in engineered viruses that lacked the gene for either glycoprotein B or Us9. Such genetically altered viruses are important as research tools, Enquist said.

Finding a way to block the activity of the proteins could be a useful strategy for treating the pain and itching associated with herpes viral diseases, Enquist said. “If you could block fusion pore formation, you could stop the generation of the signal that is causing pain and discomfort,” he said.

Granstedt conducted the experiments with Jens-Bernhard Bosse, a postdoctoral research associate in molecular biology. Assistance with 2-photon microscopy was provided by Stephan Thiberge, director of the Bezos Center for Neural Circuit Dynamics at the Princeton Neuroscience Institute.

The team previously observed the synchronized firing in laboratory-grown neurons (PLoS Pathogens, 2009), but the new study expands on the previous work by observing the process in live mice and including the contribution of Us9, Granstedt said.

Shingles, which is caused by the virus herpes zoster and results in a painful rash, will afflict almost one out of three people in the United States over their lifetime. Genital herpes, which is caused by herpes simplex virus-2, affects about one out of six people ages 14 to 49 years in the United States, according the Centers for Disease Control and Prevention.

This research was funded by National Institutes of Health (NIH) Grants NS033506 and NS060699. The Imaging Core Facility at the Lewis-Sigler Institute is funded by NIH National Institute of General Medical Sciences Center Grant PM50 GM071508.

Read the abstract

Granstedt, Andréa E., Jens B. Bosse, Stephan Y. Thiberge, and Lynn W. Enquist. 2013. In vivo imaging of alphaherpesvirus infection reveals synchronized activity dependent on axonal sorting of viral proteins. PNAS 2013 ; published ahead of print August 26, 2013, doi:10.1073/pnas.1311062110

Princeton researchers use mobile phones to measure happiness (Demography)

By Tara Thean, Science-Writing Intern, Office of the Dean for Research

World map

Locations of study subjects on world map (Source: Demography)

Researchers at Princeton University are developing ways to use mobile phones to explore how one’s environment influences one’s sense of well-being.

In a study involving volunteers who agreed to provide information about their feelings and locations, the researchers found that cell phones can efficiently capture information that is otherwise difficult to record, given today’s on-the-go lifestyle. This is important, according to the researchers, because feelings recorded “in the moment” are likely to be more accurate than feelings jotted down after the fact.

To conduct the study, the team created an application for the Android operating system that documented each person’s location and periodically sent the question, “How happy are you?”

The investigators invited people to download the app, and over a three-week period, collected information from 270 volunteers in 13 countries who were asked to rate their happiness on a scale of 0 to 5. From the information collected, the researchers created and fine-tuned methods that could lead to a better understanding of how our environments influence emotional well-being. The study was published in the June issue of Demography.

The mobile phone method could help overcome some of the limitations that come with surveys conducted at people’s homes, according to the researchers. Census measurements tie people to specific areas — the census tracts in which they live — that are usually not the only areas that people actually frequent.

“People spend a significant amount of time outside their census tracks,” said John Palmer, a graduate student in the Woodrow Wilson School of Public and International Affairs and the paper’s lead author. “If we want to get more precise findings of contextual measurements we need to use techniques like this.”

Palmer teamed up with Thomas Espenshade, professor of sociology emeritus, and Frederic Bartumeus, a specialist in movement ecology at the Center for Advanced Studies of Blanes in Spain, along with Princeton’s Chang Chung, a statistical programmer and data archivist in the Office of Population Research; Necati Ozgencil, a former Professional Specialist at Princeton; and Kathleen Li, who earned her undergraduate degree in computer science from Princeton in 2010, to design the free, open source application for the Android platform that would record participants’ locations at various intervals based on either GPS satellites or cellular tower signals.

Though many of the volunteers lived in the United States, some were in Australia, Canada, China, France, Germany, Israel, Japan, Norway, South Korea, Spain, Sweden and the United Kingdom.

Palmer noted that the team’s focus at this stage was not on generalizable conclusions about the link between environment and happiness, but rather on learning more about the mobile phone’s capabilities for data collection. “I’d be hesitant to try to extend our substantive findings beyond those people who volunteered.” he said.

However, the team did obtain some preliminary results regarding happiness: for example, male subjects tended to describe themselves as less happy when they were further from their homes, whereas females did not demonstrate a particular trend with regards to emotions and distance.

“One of the limitations of the study is that it is not representative of all people,” Palmer said. Participants had to have smartphones and be Internet users. It is also possible that people who were happy were more likely to respond to the survey. However, Palmer said, the study demonstrates the potential for mobile phone research to reach groups of people that may be less accessible by paper surveys or interviews.

Palmer’s doctoral dissertation will expand on this research, and his adviser Marta Tienda, the Maurice P. During Professor in Demographic Studies, said she was excited to see how it will impact the academic community. “His applied research promises to redefine how social scientists understand intergroup relations on many levels,” she said.

This study involved contributions from the Center for Information Technology Policy at Princeton University, with institutional support from the National Institutes of Health Training Grant T32HD07163 and Infrastructure Grant R24HD047879.

Read the abstract.

Palmer, John R. B., Thomas J. Espenshade, Frederic Bartumeus, Chang Y. Chung, Necati Ercan Ozgencil and Kathleen Li. 2013. New Approaches to Human Mobility: Using Mobile Phones for Demographic Research. Demography 50:1105–1128. DOI 10.1007/s13524-012-0175-z

How will crops fare under climate change? Depends on how you ask (Global Change Biology)

Research image

Mechanistic (top row) and empirical (bottom row) simulations compared recent, or “baseline,” maize production in South Africa (1979-99) to projected future production under climate change (2046-65). While both models showed a reduction in output, the third column shows that the empirical model estimated a widespread yield loss of around 10 percent (in yellow), while the mechanistic model showed several areas of increased production (in green). (Image by Lyndon Estes)

Research image 2

For wheat, the mechanistic model (top row) projected greater wheat yields, while the empirical model (bottom row) suggested that wheat-growing areas would expand by almost 50 percent. (Image by Lyndon Estes)

By Morgan Kelly, Office of Communications

The damage scientists expect climate change to do to crop yields can differ greatly depending on which type of model was used to make those projections, according to research based at Princeton University. The problem is that the most dire scenarios can loom large in the minds of the public and policymakers, yet neither audience is usually aware of how the model itself influenced the outcome, the researchers said.

The report in the journal Global Change Biology is one of the first to compare the agricultural projections generated by empirical models — which rely largely on field observations — to those by mechanistic models, which draw on an understanding of how crop growth and development are affected by the environment. Building on similar studies from ecology, the researchers found yet more evidence that empirical models may show greater losses as a result of climate change, while mechanistic models may be overly optimistic.

The researchers ran an empirical and a mechanistic model to see how maize and wheat crops in South Africa — the world’s ninth largest maize producer, and sub-Saharan Africa’s second largest source of wheat — would fare under climate change in the years 2046 to 2065. Under the hotter, wetter conditions projected by the climate scenarios they used, the empirical model estimated that maize production could drop by 3.6 percent, while wheat output could increase by 6.2 percent. Meanwhile, the mechanistic model calculated that maize and wheat yields might go up by 6.5 and 15.2 percent, respectively.

In addition, the empirical model estimated that suitable land for growing wheat would drop by 10 percent, while the mechanistic model found that it would expand by 9 percent. The empirical model projected a 48 percent expansion in wheat-growing areas, but the mechanistic reported only 20 percent growth. In regions where the two models overlapped, the empirical model showed declining yields while the mechanistic model showed increases. These wheat models were less accurate, but still indicative of the vastly different estimates empirical and mechanistic can produce, the researchers wrote.

Disparities such as these aren’t just a concern for climate-change researchers, said first author Lyndon Estes, an associate research scholar in the Program in Science, Technology and Environmental Policy in Princeton’s Woodrow Wilson School of Public and International Affairs. Impact projections are crucial as people and governments work to understand and address climate change, but it also is important that people understand how they are generated and the biases inherent in them, Estes said. The researchers cite previous studies that suggest climate change will reduce South African maize and wheat yields by 28 to 30 percent — according to empirical studies. Mechanistic models project a more modest 10 to 19 percent loss. What’s a farmer or government minister to believe?

“A yield projection based only on empirical models is likely to show larger yield losses than one made only with mechanistic models. Neither should be considered more right or wrong, but people should be aware of these differences,” Estes said. “People who are interested in climate-change science should be aware of all the sources of uncertainty inherent in projections, and should be aware that scenarios based on a single model — or single class of models — are not accounting for one of the major sources of uncertainty.”

The researchers’ work relates to a broader effort in recent years to examine the biases introduced into climate estimates by the models and data scientists use, Estes said. For instance, a paper posted Aug. 7 by Global Change Biology — and includes second author and 2011 Princeton graduate Ryan Huynh — challenges predictions that higher global temperatures will result in the widespread extinction of cold-blooded forest creatures, particularly lizards. These researchers say that a finer temperature scale than existing projections use suggests that many cold-blooded species would indeed thrive on a hotter Earth.

Scientists are aware of the differences between empirical and mechanistic models, said Estes, who was prompted by a similar comparison that showed an empirical-mechanistic divergence in tree-growth models. Yet, only one empirical-to-mechanistic comparison (of which Estes also was first author) has been published in relation to agriculture — and it didn’t even examine the impact of climate change.

The solution would be to use both model classes so that researchers could identify each class’s biases and correct for it, Estes said. Each model has different strengths and weaknesses that can be complementary when combined.

Simply put, empirical models are built by finding the relationship between observed crop yields and historical environmental conditions, while mechanistic models are built on the physiological understanding of how the plant grows and reproduces in response to a range of conditions. Empirical models, which are simpler and require fewer inputs, are a staple in studying the possible effects of climate change on ecological systems, where the data and knowledge about most species is largely unavailable. Mechanistic models are more common in studying agriculture because there is a much greater wealth of data and knowledge that has accumulated over several thousand years of agricultural development, Estes said.

“These two model classes characterize different portions of the environmental space, or niche, that crops and other species occupy,” Estes said. “Using them together gives us a better sense of the range of uncertainty in the projections and where the errors and limitations are in the data and models. Because the two model classes have such different structures and assumptions, they also can improve our confidence in scenarios where their findings agree.”

Read the abstract.

Estes, Lyndon D., Hein Beukes, Bethany A. Bradley, Stephanie R. Debats, Michael Oppenheimer, Alex C. Ruane, Roland Schulze and Mark Tadross. 2013. Projected climate impacts to South African maize and wheat production in 2055: A comparison of empirical and mechanistic modeling approaches. Global Change Biology. Accepted, unedited article first published online: July 17, 2013. DOI: 10.1111/gcb.12325

The work was funded by the Princeton Environmental Institute‘s Grand Challenges Program.