Progress 02/15/12 to 02/14/13
Outputs
Target Audience: The main target audience are scientists working on food policy analysis. Changes/Problems: Several factors have delayed the progress of the project including: 1) Maternity and sick leave of the Co-PI Tullaya Boonsaeng. 2) Maternity and sick leave of the Co-Pi Carlos Carpio 3) Internal re-organization within the University (the 2 Co-PIs have been in three different departments in the last 2 years. Therefore, we are now approximately 12 months behind schedule. We plan to request a 1 year no-cost project extension. What opportunities for training and professional development has the project provided? Two graduate students were involved in the project. They both became familiarized with data management techniques of big datasets as well as with the technical aspects of consumer demand estimation. How have the results been disseminated to communities of interest? We presented two papers at the Annual Meetings of the American Agricultural Economics Association which is attended by professionals in the field. Each session was attended by approximately 30 scientists. The papers are publicly available in the Ageconsearch website. According to the website statistics, the papers have been download a total of 179 times during the reporting period. What do you plan to do during the next reporting period to accomplish the goals? During the next year we plan to finish Phase 1 and start working on Phase 2 of the project.
Impacts
What was accomplished under these goals?
The efforts in year 2 were basically focused on writing the results of all the analyses performed in year 1: 1) Estimation of a demand system using eight food commodity CEX expenditure data and BLS CPI based SL price using Hoderlein and Mihaleva (2008,J.Econometrics) procedure. The procedures proposed by these authors were modified to account for the presence of zero expenditures in the CEX survey data. We explored the robustness of the elasticity estimates to the use of different CPIs (monthly, quarterly and constant prices). 2) Estimation of a demand system for the same group of goods and the same demand system used in Step 1 but using Homescan data.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2012
Citation:
Brady, K., C.E., Carpio, and T. Boonsaeng: Temporal Aggregation in the Estimation of Food Demand using Cross Sectional Data : Annual Meetings of the American Agricultural and Applied Economics Association, Seattle, Washington, August 12-14, 2012.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2012
Citation:
Castellon, C., T., Boonsaeng, and C.E. Carpio: Demand System Estimation in the Absence of Price Data : Annual Meetings of the American Agricultural and Applied Economics Association, Seattle, Washington, August 12-14, 2012.
- Type:
Theses/Dissertations
Status:
Accepted
Year Published:
2012
Citation:
Leffler, K. 2012. Temporal Aggregation and Treatment of Zero Dependent Variables in the Estimation of Food Demand using Cross-Sectional Data. Unpublished M.S. thesis. Department of Applied Economics and Statistics, Clemson University.
- Type:
Theses/Dissertations
Status:
Accepted
Year Published:
2012
Citation:
Castellon, C. 2012. Demand for Food In Ecuador and the United States: Evidence from Household-Level Survey Data. Unpublished M.S. thesis. Department of Applied Economics and Statistics, Clemson University.
|
Progress 02/15/11 to 02/14/12
Outputs
OUTPUTS: The main objective of this study is to evaluate the potential of using publicly available datasets from the Bureau of Labor and Statistics (BLS) and state of the art econometric methods in lieu of the privately owned Homescan data for the econometric estimation of food demand models. Pre-estimation stage (datasets creation) 1)BLS data: We constructed two datasets using the Consumer Expenditure Survey (CEX) data and the monthly and quarterly consumer price indices (CPIs) from year 2002 to 2006 (6,000 observations/year). The first dataset corresponds to expenditures and prices of eight aggregate food commodities. The second dataset corresponds to expenditures and prices of a disaggregate fruits and vegetables group. Commodity group prices (Stone-Lewbel prices, SL price indices) were calculated using CPIs and subgroup budget shares. 2)Nielsen Homescan Data: Two datasets comparable to those constructed with the BLS data were also constructed using the Nielsen data. To construct food commodities, individual products (at the brand-flavor-size level) first had to be aggregated into aggregate products. We included Nielsen Homescan surveys for 2002-2006 (7,000 observations/year). Fisher price indices were used as estimates of commodity group prices. Estimation Stage-Phase 1 We have conducted the following analyses: 1)Estimation of a demand system using eight food commodity CEX expenditure data and BLS CPI based SL price using Hoderlein and Mihaleva (2008,J.Econometrics) procedure. The procedures proposed by these authors were modified to account for the presence of zero expenditures in the CEX survey data. We explored the robustness of the elasticity estimates to the use of different CPIs (monthly, quarterly and constant prices). 2) Estimation of a demand system for the same group of goods and the same demand system used in Step 1 but using Homescan data. Step 3 involves the comparison of elasticity and welfare measures obtained in Steps 1 and 2. However, since the time frame for data collection in the CEX survey (2 weeks) and Homescan (at least 10 months) is different, in Step 2 we estimated and compared demand systems using two different levels of temporal aggregation for each individual in the sample: a randomly selected month and the average month within a year. Given the presence of a high percentage of zero expenditure in the monthly data we used two econometric methods for estimation of demand models with this dataset: Shonkwiler and Yen (1999, AJAE) two step censored demand estimation procedure, and a simple OLS method (Blundell and Meghir, 1987, J.Econometrics). Since the monthly data is a random sample from the annual data, elasticities and marginal using the annual data are assumed to be the "true" values. Conferences, Collaborations, Dissemination: "Using Scanner Data to Answer Food Policy Questions", conference organized by the USDA-ERS June 1-2, 2011. Collaboration with USDA-ERS personal in charge of developing the QFAHPD. Two paper proposals were submitted to the 2012 American Agricultural Economics Association organizing committee. Students involved in project: Kristyn Leffler, MSc. in Applied Economics and Statistics (May 2012). PARTICIPANTS: Carlos, C.E.: Coordinate and project management, leader on econometric aspects of project and preparation of reports. Boonsaeng, T.: Oversee and work managing the data and model estimation. Graduate students working in project(M.Sc. students): 1)Kristyn Leffler: Nielsen Homescann data management and analysis. 2) Cesar Castellon: BLS data management and analysis. Partner Organizations: USDA-Economic Research Service. TARGET AUDIENCES: Nothing significant to report during this reporting period. PROJECT MODIFICATIONS: Nothing significant to report during this reporting period.
Impacts
Since the main findings of this research project concern the results of the estimation of food demand systems using different data sources, the summary of the outcomes is separated by the type of data used in the estimation. 1)Demand System Estimation using CEX expenditure data and BLS CPI based SL prices (eight food commodity groups). 1.1.A Method for Estimation of SL Prices with Censored Expenditure Data. As mentioned previously, the construction of SL price indices combines CPIs and budget shares of the sub-groups that make up the aggregate commodity. However, when expenditures in one or several of the sub-groups are zero, the aggregate commodity SL price is undefined. While Hoderlein and Mihaleva (2008) avoided the problem by dropping the censored observations, this solution, though plausible for lower levels of censoring, results quite restrictive for higher levels of zero observations. With this in mind, we propose the use of a regression imputation approach similar to the one used in the literature for the calculation of quality adjusted unit values. The procedure involves two steps: 1) regressing SL prices of non-censored observations on a set of demographic characteristics, and 2) estimating the SL prices using the parameters of the regression obtained in step 1 and the socio-demographic characteristics of households with missing prices. 1.2.Sensitivity of Estimation Results to the Use of Different CPIs Three series of SL prices are constructed using alternative regional CPIs: monthly, quarterly, and no price variation across households. Thus, we explored the robustness of the estimation results to the chosen CPI. Our overall results indicate that elasticities and marginal effects estimates are not sensitive to the type of CPI used. We conclude that the incorporation of CPI data in the calculation of SL prices plays a limited role, thereby making possible the estimation of demand systems in the absence of price information. 2)Demand System Estimation using Nielsen Homescann Data (eight food commodity groups). We conclude that the models using monthly data closely approximate the underlying annual expenditure elasticities, but do a poor job estimating own- and -cross price elasticities and marginal effects. This finding is true for both the uncensored model (OLS model) and the censored model attempting to account for the cause of the zero expenditure. As a result, we conclude that the simplicity principle applies, at least when using Homescan data: the more complex model does not provide a significant improvement in precision. The use of shorter time frames has implications for the resulting price elasticities and marginal effects: they will be inconsistent, but not consistently biased in any direction.
Publications
- Leffler, K. 2012. Temporal Aggregation and Treatment of Zero Dependent Variables in the Estimation of Food Demand using Cross-Sectional Data. Unpublished M.S. thesis. Department of Applied Economics and Statistics, Clemson University.
|