Source: PENNSYLVANIA STATE UNIVERSITY submitted to
NUTRIENT-SPECIFIC ADDICTION IN UNHEALTHY FOODS: INVESTIGATING DYNAMIC FOOD-PURCHASING BEHAVIOR AND POLICY IMPLICATIONS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
EXTENDED
Funding Source
Reporting Frequency
Annual
Accession No.
1018747
Grant No.
2019-67023-29346
Project No.
PENW-2018-08515
Proposal No.
2018-08515
Multistate No.
(N/A)
Program Code
A1641
Project Start Date
Apr 15, 2019
Project End Date
Apr 14, 2024
Grant Year
2019
Project Director
Jaenicke, E. C.
Recipient Organization
PENNSYLVANIA STATE UNIVERSITY
408 Old Main
UNIVERSITY PARK,PA 16802-1505
Performing Department
Agricultural Economics & Rural
Non Technical Summary
As of 2012, 76 million American adults were obese, and the prevalence of this disease, which directly affects more than two-thirds of the adult population, is two-to-three times higher than it was in the 1970s. However, optimal policies to address long-standing obesity problems are not clear, with policy targets ranging from dietary fat, empty calories, and energy balance to consumer education, nutrition labeling, and even certain fat- or sugar-based taxes. These later policies emphasize the role of consumer behavior, and scientists from both social physical science fields have linked addictive behaviors with obesity.Like some previous efforts, including economic studies that capture some aspect of consumer dynamics, we intend to investigate addition both towards specific products and to ingredients or nutrients. Previous research on addictive behavior in food has captured some dynamic aspects of behavior. However, a fully dynamic model with endogenous demand that can identify both stockpiling and addictive behavior not been developed and applied towards food products. Thus, our proposed research will (a) develop a new structural model of rational addiction that is more thoroughly dynamic, (b) investigate addictive behavior at both the product level and nutrient level, (c) consider heterogeneous behavior based on consumers' health status, and (d) investigate the implications for hypothetical policies aimed at stemming the obesity epidemic in the U.S. Our project will make use of micro-level consumer purchase data linked to self-stated health outcomes.
Animal Health Component
0%
Research Effort Categories
Basic
80%
Applied
20%
Developmental
(N/A)
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
60762303010100%
Knowledge Area
607 - Consumer Economics;

Subject Of Investigation
6230 - Marketing systems and sectors thereof;

Field Of Science
3010 - Economics;
Goals / Objectives
The long-term goal of our proposed research project is to develop a deeper understanding of underlying consumer behavior that links dietary choices and health outcomes. In other words, we would like to know if some consumers are behaving as if they are addicted to certain unhealthy foods (or nutrients), and how this type of behavior interacts with other incentive-based behaviors, including general consumer response to average prices and temporary price reductions. Armed with this better understanding, we also plan to investigate how consumers of varying health status respond to a number of policy interventions, including taxes on some unhealthy projects. To achieve our long-term goal, this proposal has the following six specific objectives:Objective 1. Access and manage IRI Consumer Panel Data, IRI MedProfiler consumer health data, and Ad$pender advertising data to develop a household-specific dataset that links 2010-2017 food-purchase information to health and marketing data.Objective 2: Qualitatively evaluate the purchasing patterns across various food products, both health and unhealthy, and across nutrients for households in both the full dataset and data subsamples based on obesity status and other factors.Objective 3: Design and estimate a structural dynamic economic model of nutrient-specific food addiction with endogenous consumption and stockpiling in multiple food categories, and apply this model to consumer-level food purchase data.Objective 4: Assess the role of health information, consumer learning, advertising and promotion, and related factors in influencing the development of addiction.Objective 5: Investigate the impact of health status, including incidence of obesity, diabetes, and other chronic diseases, on addictive behavior.Objective 6: Under the framework of rational food addiction, evaluate the effectiveness of various policies aimed at reducing the consumption of unhealthy food and beverages by predicting how all consumers, as well as consumer subsamples differentiated by health status, respond to policy interventions.
Project Methods
3.1 Data Preparation, Stage 1The proposed project plans to utilize a combination of several large datasets, including (1) IRI consumer panel data, (2) IRI MedProfiler consumer health data, (3) Nexis Uni news and information data, and (4) Ad$pender advertising data. We expect to build a household-specific dataset that links 2010-2017 food-purchase information to health information/news, consumer health, and marketing data. A broad set of food categories will be chosen in this stage, including sugar-sweetened beverages, candy, salty snacks, frozen prepared food, caffeinated drinks, and similar products.The IRI consumer panel data include over 120,000 households, with 46 to 52 percent of the households providing sufficient purchase data to be included in the static panel used for analyses. We plan to collect the purchase information of all chosen food categories, including brand names, prices, package sizes, purchase quantity, and purchase date. The data also include key demographic variables including household size, age of head of household, income, ethnicity, race, children, and related characteristics. The IRI data also contain substantial calorie and macronutrient content from products' nutrition facts panel information.The IRI MedProfiler data reflect an opt-in survey on medical conditions offered to all households in the IRI consumer panel data to identify health opinions and concerns, medical conditions, and diet and lifestyle choices. About one-third of the consumer panel households had at least one member respond to the MedProfiler survey, with responses received from 95,000 to 123,000 individuals in those households. For each individual, the data contain information about the health demographics (such as age, BMI, diagnosed status of chronic diseases, etc.), health concerns and opinions, lifestyle (exercise) and diet (gluten free, low fat, vegetarian, etc.). The data can be linked with the IRI consumer panel data to provide a more complete picture of consumer health and wellness.Nexis Uni news and information data provide access to over 15,000 news, business, and legal sources, and we will consider newspapers, magazines, TV, radio, and online sources in our study.Ad$pender advertising data, obtained from Kantar Media, provide advertising expenditure data n product brands and parent companies for many major markets in the U.S. market. In addition, the advertising expenditure are classified by various types of media.3.2 Preliminary and Reduced-Form Analysis of Nutrient-Specific Addiction, Stage 2Using the prepared dataset, we plan to conduct extensive preliminary data analyses. We will first qualitatively evaluate the purchasing patterns across various food products, both healthy and unhealthy, and across nutrients for households in both the full dataset and data subsamples based on obesity status and other factors.Our reduced-form analysis is targeted for documenting descriptive evidence for (1) the nutrient-specific addictive behavior in each food category; (2) the impact of individual's health status (e.g. obesity and heart disease) on addictive behavior; (3) the impact of health news and consumer learning on addictive behavior; and (4) the impact of product advertising on addictive behavior.To empirically implement and estimate the effects of health information, for each nutrient-specific food category, we construct a two-equation model, one that specifies current consumption as a function of past and expected future consumption and prices, and a second that specifies the weight given to expected future consumption as a function of health, health news advertising, and household-specific characteristics.Even without a full set of dynamic behaviors accounted for, we expected result from these preliminary, reduced-form estimations to be interesting to peer researchers, policymakers, and industry professions. Thus, we plan to develop results from this stage into a series of outputs, e.g., presentations and manuscripts, that will be dysmyelinated to our peers and interested readers. The estimation is conducted by fixed-effects IV regression.3.3 Structural Econometric Analysis of Nutrient-Specific Addiction, Stage 3Our structural econometric analysis is aimed at addressing three related consumer-response issues. First, we propose to assess the role that information, both in terms of advertising and consumer learning, plays in influencing the development of addition. Second, we plan to investigate how consumers, and consumer subgroups differentiated by health status, respond to various policy interventions. And lastly, we propose to examine the efficacy of alternative policies targeted at reducing the consumption unhealthy foods and beverages.We build on the frameworks of Hendel and Nevo (2006), Richards, Patterson, and Tegene (2006), and Wang (2015) to estimate a dynamic demand model of rational addiction incorporating endogenous consumption and stockpiling. The advantages of this model are threefold: (a) it allows us to account for household purchases across a range of food categories, (b) it allows individual household's utility to be a function of specific nutrients, and (c) it accounts for the possibility of nutrient-specific addition and stockpiling behavior.Households in our demand model maximize the sum of discounted utilities derived from consuming their chosen products and an outside good. Similar to the model setup in Hendel and Nevo (2006) and Wang (2015), each household obtains a per period flow utility and specify a functional formed used in previous studies.A unique feature of the Hendel and Nevo (2006) model setup is the separation of the utility of consumption and the utility of purchase, which significantly decreases computational cost. We also follow Richards, Patterson, and Tegene (2006)'s model setup in implementing the Becker and Murphy (1988) rational addiction framework. More specifically, we include a parameter that measures the product-specific preferences of a household. The model allows for state-dependent preferences in the following way: It allows utility to reflect both habit persistence and rational addiction. Utility depends on the attribute-space distance of each nutrient purchased in the current period from that of the previous period and that of the following period. If either distance rises, then the household is variety-seeking. If utility falls with the distance between the current and last periods, then the household is habituated, but if the distance falls between the current and next period, then the household is thought to be rationally addicted.Once parameters of the above model are recovered, our counterfactual analysis will allow us to independently consider the three objectives listed at the beginning of this section.Using the estimated parameters above dynamic model, we will proceed to evaluate a series of policies aimed at decreasing the consumption of unhealthy food and beverages. Specifically, we plan to conduct and compare the effectiveness of the following sets of policies.(1) First, we will consider a set of tax policies on certain nutrients (sugar, salt, caffeine, etc.) that will raise retail prices in certain food categories.(2) Second, we will investigate potential policies which limit the amount of salt, sugar, or caffeine that can be added in food.(3) Third, we assess the impact of advertising restrictions in certain food categories under the framework of rational addiction.(4) Lastly, we will simulate the impact of a nation-wide public health campaign to discourage consumption of unhealthy food and beverage.

Progress 04/15/21 to 04/14/22

Outputs
Target Audience:No changes to "Target Audience". Changes/Problems:While there are no specific changes or problems affecting the current study team and project, the study team did request and receive approval for a one-year no-cost extension to the study period. The project has experiences routine logistical difficulties (and slowness) of working with very large datasets within a secure data enclave. What opportunities for training and professional development has the project provided?PhD student Josh Reed continues to meet with Jaenicke, Liu, and Wang on a weekly basis. So far, he has presented preliminary results internally at Penn State and externally to attendees at the 2021 AAEA conference (and soon the 2022 AAEA conference). How have the results been disseminated to communities of interest?Preliminary results have been presented, in the form of seminars, at several Agricultural Economics (Penn State, Arizona State, and Montana State) and at the 2021 and 2022 AAEA Conference. What do you plan to do during the next reporting period to accomplish the goals?In the next period, we plan on estimating reduced-form, rational addiction model individually for each household in the datasets, and then aggregating results to determine the percentage of households that meet the definition of rational addiction. In a second-stage analysis, we plan to investigate which household characteristics are linked to rational addiction status. For example, we will investigate if rational addiction status is linked to BMI, the presence of health conditions such as diabetes, and socioeconomic characteristics such as income, race, or age. After finding the product (or products) that exhibit the highest level of rational addiction, we will then construct a structural dynamic demand model and apply it to the data.

Impacts
What was accomplished under these goals? The study team continues work on Objectives 1, 2, and 5, and has begun work on Objective 3. Datasets for consumption of diet and regular soda, low-fat and regular milk, and potato and tortilla chips have been constructed and linked to IRI MedProfiler data so that household consumption can be linked to BMI calculations. Reduced-form models of rational addiction have been estimated for (a) each of these datasets, (b) obesity class subsets of each dataset, and (c) each individual in the dataset, separately. So far, we find only modest evidence of rational addition in these categories.

Publications

  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2022 Citation: Jaenicke, E.C., J. Reed, Y. Liu, E. Wang, and E. Zeballos, Investigating Rational Addiction To Foods At The Household Level. Invited Presentation at Arizona State University, Morrison School of Agribusiness, March 23, 2022.
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2022 Citation: Jaenicke, E.C., J. Reed, Y. Liu, E. Wang, and E. Zeballos, Investigating Rational Addiction To Foods At The Household Level. Invited Presentation at Montana State University, May 6, 2022.
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2022 Citation: Reed, J., Y. Liu, E.C. Jaenicke, E. Wang, and E. Zeballos, Food Addiction and Obesity. Selected Presentation (Accepted) at August 2022 Annual Meeting of the Agricultural and Applied Economics Association, August 1, 2022.


Progress 04/15/20 to 04/14/21

Outputs
Target Audience: Nothing Reported Changes/Problems:In general, the have been no specific changes or problems affecting the current study team and project. However, The Covid-19 situation has forced all work and meetings to be accomplished remotely, which has slowed the pace of results by a small amount. What opportunities for training and professional development has the project provided?The project team started working with a senior Ph.D. student who graduated in May 2020. In August 2020, the project team started working with a new Ph.D. student, Joshua Reed, who now has official access to the IRI data via the NORC data enclave and has been successfully completing most of the data tasks so far. He has been meeting with Jaenicke, Liu, and Wang each week since August. Josh will present the Selected Presentation, noted above, as a poster at the 2021 AAEA meetings. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period, we plan to accomplish the following: A. We plan to complete a reduced-form analysis of rational addiction (following Chaloupka, J. or Political Economy, 1991) after expanding the methods to account for multiple obesity classes of consumers. A manuscript should be completed by the end of July 2021. B. We plan to code and estimate a dynamic demand model for chips and soda that accounts both for rational addiction and stockpiling behavior. The first estimation is planned for August 2021. If the estimation is successful, we will expand this effort to additional product categories in Fall 2021 and Winter 2022

Impacts
What was accomplished under these goals? The study team continues to work mostly on Objectives 1, 2, and 5. We have now processed IRI consumer panel data, IRI retail-level data, and IRI MedProfiler data to build a panel dataset that contains multi-year, household-specific purchases of two (unhealthy) food categories - chips and soda, along with price and promotion information for all these purchases and the BMI and obesity class information for household members. Furthermore, following the methods of Gordon and Sun (Marketing Science, 2015) our qualitative analysis does support continued investigation of rational addition in the chips and soda category. Moreover, we find several important differences across obesity classes when it comes to households' purchasing trends and their responses to price promotions.

Publications

  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2021 Citation: Reed, J., E.C. Jaenicke, Y. Liu, and E. Wang, Rational Addiction and other Purchasing Dynamics Across Obesity Status Groups. Selected Presentation accepted for the August 2021 Annual Meeting of the Agricultural and Applied Economics Association.


Progress 04/15/19 to 04/14/20

Outputs
Target Audience:This project uses IRI scanner data to investigate whether food-purchase behavior across various product categories show signs of addictive behavior. Previous research on addictive behavior in food has captured some dynamic aspects of behavior. However, a fully dynamic model with endogenous demand that can identify both stockpiling and addictive behavior not been developed and applied towards food products. Thus, our proposed research will (a) develop a new structural model of rational addiction that is more thoroughly dynamic, (b) investigate addictive behavior at both the product level and nutrient level, (c) consider heterogeneous behavior based on consumers' health status, and (d) investigate the implications for hypothetical policies aimed at stemming the obesity epidemic in the U.S. Our project will make use of micro-level consumer purchase data linked to self-stated health outcomes. Changes/Problems:There have been three major personnel changes in the project, and both changes have affected the project's timeline. First, the study team had to find a replacement for our original USDA-ERS collaborator, who no longer works for USDA. The new USDA-ES collaborator is Eliana Zeballos. Second, the study team had to wait for the new collaborator to be established before successfully entering into a data agreement with USDA for full data access. That data agreement is now in place for Penn State; however, UMass also still needs a similar agreement. Third, a senior graduate student on the study team completed her dissertation at Penn State and graduated. The study team will be adding a new PhD student to the team in August 2020. In addition to these three personnel changes, two other events have slowed research progress: the move of USDA-ERS and USDA-NIFA from Washington, DC, to Kansas City, caused a temporary slowdown, and the COVID-19 situation has also caused a slowdown. What opportunities for training and professional development has the project provided?One Ph.D. student, Xuan Chen, helped with Objective 2, and used her experience on the project to help get ready for the job market. She graduated from Penn State in May 2020, and now works for Amazon. The project team will not hire a second Ph.D. student to start project work in Fall 2020. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period, we expect to work closely with our new USDA-ERS collaborator and hire a new PhD graduate assistant. Both personnel changes will allow us to begin work on Objective 3, a structural model of dynamic demand for various food categories.

Impacts
What was accomplished under these goals? The study team has mostly been working on Objective 2, the preliminary and reduced-form analysis of nutrient-specific addiction, using energy drinks as a case study. Based on the Becker and Murphy (1988) model, we have conducted a series of reduced-form analysis to test rational addiction behaviors. First, our analysis finds evidence of addiction in the energy drinks sector: consumers tend to have a higher probability of increasing purchase volume over time. Because the rational addiction model implies that addicted consumers are more likely to increase the quantities of their successive purchases, the results present a strong motivation to incorporate the addiction into purchase behaviors.Second, our analysis shows that consumers are forward-looking, not myopic, which means they link consumption today to the potential higher costs in the future and take it into consideration when making a decision. Third, we incorporate measures of individuals' access to health information into the estimation and find that an increase in health information, or learning the health risks of energy drinks, negatively influences the consumption of energy drinks. However, this effect is decreasing over time. Overall, using the reduced-form analysis and energy drinks as a test case, we find evidence of rational addiction at the aggregated level. However, the reduced-form regression cannot capture the potential heterogeneous addiction behaviors or separate the potential influence of health information from the impact of addiction. Further, it cannot capture the stockpiling features in consumer purchases. Therefore, a full dynamic structure model is needed, which will be conducted in Stage 3/Obbjective 3.

Publications

  • Type: Theses/Dissertations Status: Submitted Year Published: 2020 Citation: PhD Dissertation, Xuan Chen, Penn State Univesity