Chen Zhen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Chen Zhen has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
Biing‐Hwan Lin
U.S. Department of Agriculture Economic Research Service Washington DC USA
2 shared publicationsLisa Mancino
U.S. Department of Agriculture Economic Research Service Washington DC USA
2 shared publicationsMichele Ver Ploeg
U.S. Department of Agriculture Economic Research Service Washington DC USA
2 shared publicationsShawn Karns
RTI International Durham North Carolina USA
2 shared publicationsYu Chen
Department of Agricultural and Applied Economics University of Georgia Athens Georgia USA
2 shared publicationsAbigail Okrent
Food Economics Division, Economic Research Service, U.S. Department of Agriculture, Washington, DC
1 shared publication
Latest Publications
- 15 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 11 (3) , 1235-1246
Exploring New Frontiers in Agricultural NLP: Investigating the Potential of Large Language Models for Food Applications
This paper explores new frontiers in agricultural natural language processing (NLP) by investigating the effectiveness of food-related text corpora for pretraining transformer-based language models. Specifically, we focus on semantic matching, establishing mappings between food descriptions and nutrition data through fine-tuning AgriBERT with the FoodOn ontology. Our work introduces an expanded comparison with state-of-the-art language models such as GPT-4, Mistral-large, Claude 3 Sonnet, and Gemini 1.0 Ultra. This exploratory investigation, rather than a direct comparison, aims to understand how AgriBERT, a domain-specific, fine-tuned, open-source model, complements the broad knowledge and generative abilities of these advanced LLMs in addressing the unique challenges of the agricultural sector. We also experiment with other applications, such as cuisine prediction from ingredients, expanding our research to include various NLP tasks beyond semantic matching. Overall, this paper underscores the potential of integrating domain-specific models like AgriBERT with advanced LLMs to enhance the performance and applicability of agricultural NLP applications.
- 15 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 11 (3) , 1235-1246
Exploring New Frontiers in Agricultural NLP: Investigating the Potential of Large Language Models for Food Applications
This paper explores new frontiers in agricultural natural language processing (NLP) by investigating the effectiveness of food-related text corpora for pretraining transformer-based language models. Specifically, we focus on semantic matching, establishing mappings between food descriptions and nutrition data through fine-tuning AgriBERT with the FoodOn ontology. Our work introduces an expanded comparison with state-of-the-art language models such as GPT-4, Mistral-large, Claude 3 Sonnet, and Gemini 1.0 Ultra. This exploratory investigation, rather than a direct comparison, aims to understand how AgriBERT, a domain-specific, fine-tuned, open-source model, complements the broad knowledge and generative abilities of these advanced LLMs in addressing the unique challenges of the agricultural sector. We also experiment with other applications, such as cuisine prediction from ingredients, expanding our research to include various NLP tasks beyond semantic matching. Overall, this paper underscores the potential of integrating domain-specific models like AgriBERT with advanced LLMs to enhance the performance and applicability of agricultural NLP applications.
- Open Access
- 29 June 2023
- Published by Wiley in American Journal of Agricultural Economics
- Vol. 106 (3) , 1058-1088
Do obese and nonobese consumers respond differently to price changes? Implications of preference heterogeneity for obesity‐oriented food taxes and subsidies
Preference heterogeneity in food demand has important health and equity implications for targeted and broad‐based taxes and subsidies intended to enhance diet quality and reduce obesity. We study the role of obesity in the purchases of food at home and food away from home using data from the nationally representative National Household Food Acquisition and Purchase Survey. We develop a method for incorporating the complex survey design and retail scanner data into the estimation of a 21‐good Exact Affine Stone Index demand system with endogenous prices and truncated purchases. We find significant preference heterogeneity associated with the obesity status of household members. Counterfactual simulations find that (1) a sweetened beverage tax is effective in increasing the healthfulness of purchases by lower income obese consumers; (2) the nutritional benefits of a fruit and vegetable subsidy are concentrated on nonobese consumers with little improvement in obese consumers' Healthy Eating Index and an increase in their total calories purchased; and (3) a fiscally neutral healthy food subsidy fully funded by an unhealthy food tax benefits nonobese consumers both financially and nutritionally more than it does obese consumers. These findings show that lowering healthy food prices without raising the cost of unhealthy foods is unlikely to reduce obesity. Policymakers in favor of a systems approach of simultaneously taxing unhealthy foods and subsidizing healthy foods should be mindful of the distributional effects of this policy on obese consumers and the lower income population.
- 29 June 2023
- Published by Wiley in American Journal of Agricultural Economics
- Vol. 106 (3) , 1058-1088
Open AccessDo obese and nonobese consumers respond differently to price changes? Implications of preference heterogeneity for obesity‐oriented food taxes and subsidies
Preference heterogeneity in food demand has important health and equity implications for targeted and broad‐based taxes and subsidies intended to enhance diet quality and reduce obesity. We study the role of obesity in the purchases of food at home and food away from home using data from the nationally representative National Household Food Acquisition and Purchase Survey. We develop a method for incorporating the complex survey design and retail scanner data into the estimation of a 21‐good Exact Affine Stone Index demand system with endogenous prices and truncated purchases. We find significant preference heterogeneity associated with the obesity status of household members. Counterfactual simulations find that (1) a sweetened beverage tax is effective in increasing the healthfulness of purchases by lower income obese consumers; (2) the nutritional benefits of a fruit and vegetable subsidy are concentrated on nonobese consumers with little improvement in obese consumers' Healthy Eating Index and an increase in their total calories purchased; and (3) a fiscally neutral healthy food subsidy fully funded by an unhealthy food tax benefits nonobese consumers both financially and nutritionally more than it does obese consumers. These findings show that lowering healthy food prices without raising the cost of unhealthy foods is unlikely to reduce obesity. Policymakers in favor of a systems approach of simultaneously taxing unhealthy foods and subsidizing healthy foods should be mindful of the distributional effects of this policy on obese consumers and the lower income population.
- 28 October 2021
- Published by Wiley in Agribusiness
- Vol. 38 (1) , 3-20
The potential impact of reducing sodium in packaged food: The case of the Chinese instant noodles market
Reducing sodium intake has been identified as a priority and cost‐effective intervention in lowering the risk of hypertension and cardiovascular diseases. To accomplish the population‐wide sodium reduction, two main strategies regarding food supply have been widely discussed: one is to set mandatory standards for the sodium content of foods and the other is to call for efforts from the food industry to voluntarily reduce sodium. Using the Chinese instant noodles market as an example, we estimate a random‐coefficients logit demand model to quantify the degree of substitution between competing products and between sodium and fat. The demand estimates are then used to simulate three sodium reduction strategies: a unilateral reduction by one firm, and industry‐wide reductions by multiple firms with and without reformulation of fat content. We find that voluntarily sodium reduction by a single firm results in its sales loss to its competitors; an industry‐wide sodium reduction causes demand reduction to be more evenly distributed across participating firms; and an industry‐wide reduction accompanied by an increase in fat content largely negates the adverse effect of sodium reduction on sales but risks an increase in population fat intake. [EconLit Citations: D12; I18; L66].
- 28 October 2021
- Published by Wiley in Agribusiness
- Vol. 38 (1) , 3-20
The potential impact of reducing sodium in packaged food: The case of the Chinese instant noodles market
Reducing sodium intake has been identified as a priority and cost‐effective intervention in lowering the risk of hypertension and cardiovascular diseases. To accomplish the population‐wide sodium reduction, two main strategies regarding food supply have been widely discussed: one is to set mandatory standards for the sodium content of foods and the other is to call for efforts from the food industry to voluntarily reduce sodium. Using the Chinese instant noodles market as an example, we estimate a random‐coefficients logit demand model to quantify the degree of substitution between competing products and between sodium and fat. The demand estimates are then used to simulate three sodium reduction strategies: a unilateral reduction by one firm, and industry‐wide reductions by multiple firms with and without reformulation of fat content. We find that voluntarily sodium reduction by a single firm results in its sales loss to its competitors; an industry‐wide sodium reduction causes demand reduction to be more evenly distributed across participating firms; and an industry‐wide reduction accompanied by an increase in fat content largely negates the adverse effect of sodium reduction on sales but risks an increase in population fat intake. [EconLit Citations: D12; I18; L66].
- Open Access
- 18 December 2020
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 15 (12) , e0240263
Nutritional quality of retail food purchases is not associated with participation in the Supplemental Nutrition Assistance Program for nutrition-oriented households
The Supplemental Nutrition Assistance Program (SNAP) provides millions of low-income Americans food benefits and other forms of nutrition assistance. Evidence indicates that SNAP reduces food insecurity. However, there is a concern that the food benefit may increase the demand for less healthy foods more than healthier foods, thereby reducing the overall nutritional quality of the participant’s food basket. This paper aims to examine the association of SNAP participation with the nutritional quality of food-at-home purchases of low-income households and to investigate the potential heterogeneity among consumers with different levels of nutrition attitude. This analysis used food purchase data from the USDA National Household Food Acquisition and Purchase Survey (FoodAPS). Our study sample included 2,218 low-income households, of which 1,184 are SNAP participants, and 1,034 are income-eligible nonparticipants. Multivariate regressions were performed to explore the SNAP-nutritional quality association. A household’s nutrition attitude was measured using its response to a question on whether the household searched for nutrition information online in the last 2 months. Households that affirmed they had an online nutrition search were treated as nutrition-oriented households (21.2% of the low-income sample), and households that did not were considered less nutrition-oriented households (78.8%). For robustness, we also created an alternative nutrition attitude measure based on reported use of the nutrition facts label. We found that among less nutrition-oriented households, SNAP participants had a statistically significant 0.097 points (p = 0.018) lower Guiding Stars rating than low-income nonparticipants. However, there was no significant SNAP-nutritional quality association among nutrition-oriented households. In conclusion, SNAP participation was associated with lower nutritional quality of food purchases among less nutrition-oriented households, but not among nutrition-oriented households. The results suggest that the intended nutritional benefits of restrictions on purchases of healthy foods may not reach the subgroup of nutrition-oriented SNAP participants.
- 18 December 2020
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 15 (12) , e0240263
Open AccessNutritional quality of retail food purchases is not associated with participation in the Supplemental Nutrition Assistance Program for nutrition-oriented households
The Supplemental Nutrition Assistance Program (SNAP) provides millions of low-income Americans food benefits and other forms of nutrition assistance. Evidence indicates that SNAP reduces food insecurity. However, there is a concern that the food benefit may increase the demand for less healthy foods more than healthier foods, thereby reducing the overall nutritional quality of the participant’s food basket. This paper aims to examine the association of SNAP participation with the nutritional quality of food-at-home purchases of low-income households and to investigate the potential heterogeneity among consumers with different levels of nutrition attitude. This analysis used food purchase data from the USDA National Household Food Acquisition and Purchase Survey (FoodAPS). Our study sample included 2,218 low-income households, of which 1,184 are SNAP participants, and 1,034 are income-eligible nonparticipants. Multivariate regressions were performed to explore the SNAP-nutritional quality association. A household’s nutrition attitude was measured using its response to a question on whether the household searched for nutrition information online in the last 2 months. Households that affirmed they had an online nutrition search were treated as nutrition-oriented households (21.2% of the low-income sample), and households that did not were considered less nutrition-oriented households (78.8%). For robustness, we also created an alternative nutrition attitude measure based on reported use of the nutrition facts label. We found that among less nutrition-oriented households, SNAP participants had a statistically significant 0.097 points (p = 0.018) lower Guiding Stars rating than low-income nonparticipants. However, there was no significant SNAP-nutritional quality association among nutrition-oriented households. In conclusion, SNAP participation was associated with lower nutritional quality of food purchases among less nutrition-oriented households, but not among nutrition-oriented households. The results suggest that the intended nutritional benefits of restrictions on purchases of healthy foods may not reach the subgroup of nutrition-oriented SNAP participants.
- 1 December 2019
- Published by Elsevier in Economics & Human Biology
- Vol. 35, 133-143
Does point-of-sale nutrition information improve the nutritional quality of food choices?
- 1 December 2019
- Published by Elsevier in Economics & Human Biology
- Vol. 35, 133-143
Does point-of-sale nutrition information improve the nutritional quality of food choices?
- 22 May 2019
- Published by Wiley in Health Economics
- Vol. 28 (6) , 782-800
Do differences in reported expenditures between household scanner data and expenditure surveys matter in health policy research?
Household scanner data are increasingly used to inform health policy such as sugar‐sweetened beverage taxes. This article examines whether differences in the level of reported expenditures between IRI Consumer Network scanner panel and the Consumer Expenditure Survey (CES) lead to important differences in demand elasticities and policy simulation outcomes. Using each dataset, we estimated a structural consumer demand system with seven food groups and a numéraire good. To compare the two datasets on a level playing field, we went to great lengths to ensure that the explanatory variables in the two demand models were comparably constructed. Results indicate that scanner data households are not consistently more price responsive than the general population and underreported Consumer Network expenditures do not seem to result in systematic differences in price elasticities. The income elasticities are uniformly lower in Consumer Network than in CES for higher income households because of the positive association between income and the degree of underreporting. This, however, has limited effects on uncompensated price elasticities and policy simulations because food budget shares are small for higher income households. Overall, these findings support continued use of household scanner data in health policy research related to effects of price (dis)incentives.
- 22 May 2019
- Published by Wiley in Health Economics
- Vol. 28 (6) , 782-800
Do differences in reported expenditures between household scanner data and expenditure surveys matter in health policy research?
Household scanner data are increasingly used to inform health policy such as sugar‐sweetened beverage taxes. This article examines whether differences in the level of reported expenditures between IRI Consumer Network scanner panel and the Consumer Expenditure Survey (CES) lead to important differences in demand elasticities and policy simulation outcomes. Using each dataset, we estimated a structural consumer demand system with seven food groups and a numéraire good. To compare the two datasets on a level playing field, we went to great lengths to ensure that the explanatory variables in the two demand models were comparably constructed. Results indicate that scanner data households are not consistently more price responsive than the general population and underreported Consumer Network expenditures do not seem to result in systematic differences in price elasticities. The income elasticities are uniformly lower in Consumer Network than in CES for higher income households because of the positive association between income and the degree of underreporting. This, however, has limited effects on uncompensated price elasticities and policy simulations because food budget shares are small for higher income households. Overall, these findings support continued use of household scanner data in health policy research related to effects of price (dis)incentives.
- 18 June 2018
- Published by Wiley in American Journal of Agricultural Economics
- Vol. 101 (1) , 311-329
Scanner Data‐Based Panel Price Indexes
We construct panel price indexes using retail scanner data that allow comparisons of consumption cost across space and time. Two types of panel indexes are examined: the rolling‐window panel extensions of the multilateral Cave‐Christensen‐Diewert index with the Törnqvist index as its elements, and of the multilateral Gini‐Eltetö‐Köves‐Szulc index using the Fisher ideal index as its elements. The rolling window method maintains the nonrevisability of published index numbers while allowing index numbers for new periods and locations to be calculated and the basket of items to be updated. Meanwhile, the multilateral structure of price comparison eliminates significant downward drift in standard chained indexes. Using county‐level bilateral and panel indexes based on retail beverage scanner data, we experimentally adjust for purchasing parity the portion of Supplemental Nutrition Assistance Program (SNAP) benefits that participants spend on beverages. Accounting for temporal and spatial cost differences causes over 2% of SNAP allotment spent on beverages to be reallocated, or approximately a 5% change in allotment on average for a county. About 90% of the relocated SNAP fund is to adjust for spatial differences in food cost. We also compare SNAP allotments implied by the retail scanner data indexes with those implied by indexes based on the USDA Quarterly Food‐at‐Home Price Database (QFAHPD). The treatment of unit values and product quality may have contributed to the significant differences observed between the retail scanner data indexes and the QFAHPD indexes.
- 18 June 2018
- Published by Wiley in American Journal of Agricultural Economics
- Vol. 101 (1) , 311-329
Scanner Data‐Based Panel Price Indexes
We construct panel price indexes using retail scanner data that allow comparisons of consumption cost across space and time. Two types of panel indexes are examined: the rolling‐window panel extensions of the multilateral Cave‐Christensen‐Diewert index with the Törnqvist index as its elements, and of the multilateral Gini‐Eltetö‐Köves‐Szulc index using the Fisher ideal index as its elements. The rolling window method maintains the nonrevisability of published index numbers while allowing index numbers for new periods and locations to be calculated and the basket of items to be updated. Meanwhile, the multilateral structure of price comparison eliminates significant downward drift in standard chained indexes. Using county‐level bilateral and panel indexes based on retail beverage scanner data, we experimentally adjust for purchasing parity the portion of Supplemental Nutrition Assistance Program (SNAP) benefits that participants spend on beverages. Accounting for temporal and spatial cost differences causes over 2% of SNAP allotment spent on beverages to be reallocated, or approximately a 5% change in allotment on average for a county. About 90% of the relocated SNAP fund is to adjust for spatial differences in food cost. We also compare SNAP allotments implied by the retail scanner data indexes with those implied by indexes based on the USDA Quarterly Food‐at‐Home Price Database (QFAHPD). The treatment of unit values and product quality may have contributed to the significant differences observed between the retail scanner data indexes and the QFAHPD indexes.