Consumer Attitudes Towards Food Safety, Trust and Meat Purchase Behaviour in Canada-A BSE-Related Study
Jun Yang1; Ellen Goddard2
1. PhD Candidate (jy8@ualberta.ca); 2. Professor, Dept. of Rural Economy, University of Alberta
Issues
Consumers make their purchase decisions based on various characteristics of products such as search and experience attributes (Nelson, 1970, 1974) as well as credence attributes (Darby and Karni, 1973). The search and experience attributes of meats can be observed such as texture, color, flavour and tenderness while the credence attributes of meats can not be observed ( meat safety and meat production practices). Meat safety may be a very important credence attribute for Canadian consumers given BSE and other food ‘scares’ and concerns about animal production practices. Attitudes toward meat consumption and trust in government, scientists and food industry can also affect purchase decisions (Yang and Li, 2004, Yang and Goddard, 2007, Steiner and Yang, 2007). The objective of this research is to investigate the role of food safety concerns, trust in various agents in the food industry on household level revealed preference meat purchase data.
Categorical Principle Component Analysis and Cluster Analysis
Categorical principle component analysis (CPCA) is a technique to explain most of the variation in observed variables by fewer underlying components. A loss function can be minimized to maximize the variation explained by underlying components.
Loss ( CS , CL , X ) = N
−1
Demand Analysis
A two-stage TransLog demand system is estimated separately over the five groups clustered. The model specifications are as follows:
First Stage:
LnTEXP = a0 +
∑
3
quarter
i =1
i
+ a 4 * ∑ w i * LnP i + a 5 * PDI + a 6 * LnTEXP
i =1
4
−1
+ a7 * t + ε 0
bi , 0 +
∑∑
i =1
N
M
∑
j =1 i, j
3
bi , j * quarter j +
4
∑
j =1 4
4
c i , j * LnP j + d i * LnTEXP + ei * LnQ i , −1 + f i * t * LnP j +
j =1
( x ij − cs iT cl j ) 2
Second Stage:
wi =
∑b
i =1
4
i ,0
+
∑∑b
i =1 j =1
4
3
* quarter j +
∑∑c
i =1 j =1
i, j
∑d
i =1
4
i
* LnTEXP +
∑ e * LnQ
i i =1
4
i , −1
+
∑f
i =1
4
+ εi
i
*t
where Loss is the loss function to be minimized. CS and CL are component scores and component loadings. N is the number of observations and M is the number of variables. X is the optimal scaling of observed categorical data matrix. The variables from questions about risk perceptions and attitude toward BSE, animal production-related concern, food and beef consumption attitude, trust to government, scientists and food industry are reduced to two dimensions.
Figure 1. Component Scores from CPCA
Risk Perceptions:
Pd beef, t = what beef 1 − what beef 2
Pd beef, t = ΔRtp + ε t = −(1 − ω ) R0 + ω * ΔRtp 1 + θ * ∑ φi * Media i + ε t −
i =1 N
Methodology
The empirical methods include categorical principal component analysis (CPCA), cluster analysis (CA) and meat demand analysis. First, based on a 2008 consumer survey, a CPCA is employed to examine the responses to questions about risk perceptions, attitudes toward BSE, animal production-related concerns, food and beef consumption attitudes, trust in government, scientists and food industry. Then, sample Canadian consumers are classified by a CA into five groups based on their factor scores from CPCA. These different groups are evaluated separately with a meat demand system including beef, pork, chicken, turkey and seafood. The price and substitution elasticities will be calculated in each group and compared. Further, based on Social Amplification of Risk Framework (SARF) and Prospective Reference Theory (PRT), an equation about consumer risk perceptions related to BSE is constructed and estimated for each group from 2004 to 2007. The application of SARF and PRT in demand analysis is an initial effort to both track consumer risk perceptions over time and to measure the impacts of quantity and quality of media information on consumer risk perceptions.
‐
Figure 2. Component Loadings from CPCA
where TEXP is the total meat expenditure per capita. Quarteri (i=1,2,3) is the quarterly dummies. a0 and bi,0 are individual-specific constants affected by demographic variables of individuals. wi is the expenditure share for meat i(i=beef, pork, chicken, turkey). Pi is the retail price of meat i. PDI is the per capita disposable income. t is the time trend. Qi,-1 is the lagged quantity of consumption for meat i to capture the habit effects. Linear homogeneity and homothetic separability are imposed under the requirement of multi-stage demand system and price aggregations. Pdbeef,t is the difference of predicted beef shares after BSE outbreak based on the data before BSE and after BSE, which is used to approximate risk perceptions over BSE (∆Rtp). Mediai is ith media information index which can be created either from quantity or quality of information.
Table 1. Own‐price Elasticities across two stages after BSE in 2003
Meat type Beef Group1 Group2 Group3 Group4 Group5 Variables Lagged risk perception over BSE Cumulative BSE information Cumulative BSE information from local media Chicken -1.01(0.12)*** -0.99(0.22)*** -0.95(0.09)*** -0.89(0.08)*** -1.23(0.06)*** Cumulative BSE information addressing government Cumulative BSE information addressing scientists Cumulative BSE information addressing devast ated farmers Time trend
Table 2. Estimated Risk Perception Equationsa
Group1 0.523(0.008)*** -0.013(0.003)*** 0.027(0.004)*** -0.011(0.001)*** 0.005(0.001)*** -0.009(0.005) -0.033(0.006)*** Group2 0.39(0.013)*** 0.001(0.001) 0.004(0.001)** 0.001(0.001) 0.001(0.001) -0.001(0.002) n/a Group3 0.34(0.008)*** 0.002(0.001) -0.002(0.002) 0.001(0.001)* 0.001(0.001) 0.001(0.003) 0.001(0.003) Group4 0.451(0.006)*** -0.004(0.001)*** 0.016(0.002)*** -0.003(0.001)*** 0.002(0.001)*** -0.008(0.002)*** -0.011(0.003)*** Group5 0.572(0.006)*** -0.001(0.003) 0.028(0.004)*** -0.006(0.001)*** 0.002(0.001) -0.008(0.005) -0.013(0.006)*
-1.1(0.16)***
-0.81(0.28)***
-1.02(0.11)***
-0.78(0.1)***
0.25(0.11)**
Pork
-1.07(0.07)***
-0.62(0.13)***
-1.14(0.05)***
-1.2(0.05)***
-0.94(0.12)***
Hierarchical cluster analysis (HCA) is applied to the two dimensional variables generated from CPCA. Five groups are formed including two groups with high levels of risk perceptions and attitude and low levels of trust (group 1 and group 2), two groups with low levels of risk perceptions and attitude and high levels of trust (group 4 and group 5) and one neutral group (group 3).
Figure 3. Percentages of Categories within Each Group
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 6 1 2 3 4 5 6
Turkey
-1.94(0.44)***
-0.82(0.85)
-1.31(0.31)***
-1.05(0.27)***
-0.98(0.1)***
Seafood
-0.49(0.14)***
-0.23(0.22)
-0.65(0.09)***
0.04(0.09)
-1.1(0.31)***
a: time trend is removed in risk perception equation of group2 to get a converge result.
Data
Survey data based on a sub-sample of ACNielsen HomeScan™ households are collected (households for whom purchases data is available from 2002 to 2007). 50% of sample households are from Ontario and Quebec. 37% from Alberta, B.C., Manitoba and Saskatchewan. The households with single or two members account for 71% of total households. Households with no children under 18 are dominant in the sample (77%). 46% of sample households have annual incomes beyond $50,000. Household heads have various education levels from high school to university. Around 60% of the households are from urban areas. Households show that 70% of their meat purchases are beef, 63% are chicken, 50% are pork and 15% are seafood.
As shown in Table 1, the elasticities in different groups suggest that consumers in group 1 who have low levels of trust and high levels of risk perceptions about BSE have a larger own-price elasticity (beef) than Group 4, for example, a group who has high levels of trust and low levels of risk perception. Group 1’s own beef price elasticity is not too dissimilar from the neutral group.
Figure 4. Estimated Risk Perceptions about BSE
Estim ated Risk Perceptions 0.25 0.2 0.15 0.1 0.05 0 2004 2005 2006 2007 Year
Percentages of Categories within Each Group
Group1
Group2
Group3
Group4
Group5
Categories Variables
Confidence in food Confidence in beef safety safety
Trust in manufacturers
Trust in retailers
Trust in government
Trust in farmers
Risk perceptions Concern over BSE Acquired media Risk perceptions over Impact of BSE in risk of eating beef and vCJD information about BSE perceptions of food BSE and vCJD safety Group 3 Group 4 Group 5
Group 1
Group 2
Categories 1 to 5 represent the extent from low to high or from small to large. As shown in Figure 3, although almost all consumers got same amount of information about BSE and vCJD, they have different attitudes and perceptions about food and beef safety. Further, consumers with low confidence in food and beef safety have low levels of trust in government, manufacturers, retailers or farmers and high risk perceptions about beef eating and BSE and vice versa.
In terms of the estimated risk perceptions about BSE from 2004 to 2007, as shown in Figure 4, group 1 has the highest risk perceptions about BSE. Group 3 as the group with neutral attitude has close-to-zero risk perceptions about BSE. Other groups have relatively modest levels of risk perceptions about BSE . The risk perception equation is estimated for each group (Table 2) and both the quantity and quality (information sources and addressed subjects) of media information have significant impacts on risk perceptions of BSE. Lagged risk perceptions about BSE also have significant impacts on current risk perceptions, suggesting a Bayesian updating of risk perceptions of BSE over time.
Conclusions
This study systematically analyzed the relationship among consumer risk perceptions, risk attitudes and trust in different organizations (government, manufacturers, retailers and farmers). The converse relations between the levels of trust and the levels of risk perceptions and attitude are confirmed by CPCA and CA. Further, the consumers in a low trust and high risk attitude group (group1) have higher risk perceptions and higher beef own-price elasticities than the consumers in some other groups. Based on the estimated risk perception equations, SARF is supported in which both quality and quantity of information play significant roles in risk amplification. PRT is also proved in which consumers adaptively update their risk perceptions over time.