AgBioForum
Volume 5 // Number 2 // Article 3
PDF Comment on this article Issue contents Previous article Next article
Consumer Attitudes Towards Genetic Modification, Functional Foods, and Microorganisms: A Choice Modeling Experiment for Beer
School of Agricultural & Resource Economics, University of Western Australia
A choice modeling approach was used to identify consumer preferences for various hypothetical forms of genetic modification in beer, using a sample from Western Australia. It was found that respondents were equally averse to first-generation modification in either plants or microorganisms but were willing to pay a premium for a product with positive health benefits.
Key words: genetic modification, food, health benefits, microorganisms.
Introduction

The first generation of genetically modified (GM) foods, with its focus on producer benefits, has met with considerable consumer resistance in a number of countries and, as a consequence, a significant policy response in terms of regulatory control and labeling. Some advocates of genetic modification see this is a temporary issue, driven in part by the lack of any direct consumer benefits from the new technology and in part by confused messages about potential economic and environmental impacts. They argue that once the next generation of products are available that show direct benefits to consumers (either in terms of improved qualities of foods or direct health benefits), the level of acceptance will be much higher (e.g., Gamble, Muggleston, Hedderley, Parminter, & Richardson-Harman, 2000; Schmidt, 2000).

The purpose of this research is to test this contention by using a choice modeling framework. Such a framework has been used extensively to investigate hypothetical changes in environmental and agricultural polices, and there have been some efforts to investigate attitudes towards GM foods. The preferred means of investigating preferences—through revealed preferences—is not available in circumstances where GM products cannot be freely traded (and, in large part, research is designed to investigate whether they should be). Direct experimental approaches, where real but trivial trades are made under experimental conditions, are unlikely to reveal the state of preferences which will be made in the context of total food purchases.

Choice modeling

Choice modeling has been taken up within the environmental valuation literature, where its ability to deal with extended attribute sets (including those related to product and process) give it considerable flexibility (e.g., Bennett & Blamey, 2001; Morrison, Blamey, Bennett, & Louviere, 1996; Adamowicz, Boxall, Williams, & Louviere, 1998). In the current context, a hypothetical product was devised and described to the respondent with alternative biotechnologies used at various points in the production process (see the Appendix for a copy of the survey). There have been a relatively small number of papers that have applied this technique to GM foodstuffs (e.g., Burton, Rigby, Young, & James, 2001; Donaghy, Rolfe, & Bennett, 2002; Owen, Louviere, & Clark, 2002; Baker & Burnham, 2001).

The product was beer, which would be familiar to all respondents. The first attribute specified was the form of barley. This was either conventional, or a GM barley that reduced production costs (i.e., a classic first-generation agricultural product). The second attribute specified was the yeast used in the brewing process. This was either conventional, or GM to reduce the costs of brewing, or GM to leave increased antioxidants in the beer, which would reduce cholesterol levels by 20% if consumed in moderation. The third attribute was the price of the beer, which varied across a range of A$2.00 to A$4.00.

This simple survey design gives three attributes. A modified greco-latin square was used to derive a main effects combination of attribute levels. Table 1 indicates the combination of attribute levels that constitute the 20 alternative beer types used in the survey. Where only a GM cost-reducing attribute is included in the product, the price is restricted to fall or at worst stay constant. For the health-enhancing product, both reduced and increased price levels are included.

Table 2 gives an indicative choice set. Each respondent was asked to complete ten of these sets. Each set contained the conventional beer type and two other options from the set of attribute combinations.

Table 1. The combinations of attribute levels used to generate 20 hypothetical beers.
  Yeast
Conventional GM(Cost) GM(Health)
Barley Conventional $3.00 $2.00, $2.50, $3.00 $2.00, $2.50, $3.00, $3.50, $4.00
GM(Cost) $2.00, $2.50, $3.00 $2.00, $2.50, $3.00 $2.00, $2.50, $3.00, $3.50, $4.00

Table 2. Example of a choice set.
Attribute Conventional Option 1 Option 2
Price $3.00 $2.50 $3.50
Barley Conventional Conventional GM(Cost)
Yeast Conventional GM(Cost) GM(Health)
If these three bottles of beer are the only ones available, which beer do you prefer?      

This experimental design allowed us to test a series of hypotheses: (a) the presence of first-generation genetic modification in the production process reduces the value of the product to the consumer; (b) this valuation will differ depending on the vehicle for the first generation process—plant or microorganism—with an expectation that microorganisms may be of less concern; (c) values with respect to first-generation genetic modification are cumulative—i.e., the greater the extent of genetic modification used in the process, the greater the concern; and (d) aversion to GM processes will be moderated if the modification generates health benefits.

Statistical analysis of the choices used the random utility model. Assume that the utility derived by individual i from product j is given by

Uij = ΣkΒkXkj + εj

where Xk are the quantified attributes of the product. If the individual selected the product that gives the highest utility, and assuming independent and identically distributed error terms following a Weibull distribution (McFadden, 1973), the probability of choosing option j from N options can be expressed as:

This is a conditional logit model, which can be estimated using a wide variety of standard statistical packages (Greene, 1997).

Parameter estimates from the conditional logit model identify the utility parameters (Louviere, Hensher, & Swait, 2000, p.39), and in the case of a linear utility function, marginal utilities. In particular, the parameter on the payment level identifies the negative of marginal utility of income. The ratio of the attribute parameters to the parameter on the payment level give partworths: the marginal $ value associated with a change in the attribute.

The survey was administered in 2001, using a drop-off/postal return with prepaid envelopes. Limited resources meant that a very limited coverage could be attained: 250 surveys were distributed in randomly-selected streets in five suburbs across Perth, Western Australia (WA). The suburbs were selected on the basis of expected income levels to get a cross section of the community. Sixty-four completed surveys were returned, for a response rate of 16%. This is not a high response rate, but was not surprising given the relative complexity of the survey and no possibility of conducting follow-up reminders. The gender balance was quite even (45% male); mean household income was $43,000, which is close to the average for couples with dependent households in WA. Median age for the sample was within 30-40 years; the median for the state was 34. The sample had a significantly higher level of educational attainment than the population average: 46% at tertiary level, compared to a national average of 27%. This may reflect a degree of self-selection when respondents were faced with a relatively complex survey instrument.

The 64 surveys provided 610 usable choices. However, 19 individuals within the set always selected the conventional beer—irrespective of the price discounts or health benefits being offered, they did not select a beer involving a GM modification. These individuals may have had a utility function consistent with the rest of the sample, but it is more likely that they had a committed opposition to GM that was not amenable to tradeoffs. This is analogous to the problem of large numbers of zero willingness-to-pay values from a conventional contingent valuation study; it may imply a subpopulation with preferences that are quite different from the rest of the population. This has been tested formally by conducting a Log Likelihood test for parameter stability, splitting the data into two sets: those respondents who showed some variation in their selection, and those who always selected the conventional beer. The results of this test suggested that the null hypothesis—that parameters were stable across the two groups—is rejected (a test statistic in excess of 400, compared to a critical value of 16.92); hence, there were two subpopulations within the sample. The remainder of the analysis focused on that group who were prepared to make a tradeoff across the attributes. However, in interpreting the total consumer response to the GM issues presented here, it should be remembered that a significant proportion of the sample did not purchase GM products for the range of prices and attributes used in the experiment.

Focusing on the set of respondents who were prepared to consider GM beer, of the individual specific characteristics collected, only two were found to be significant modifiers of attitudes towards attribute levels: the age of the respondent and whether they considered cholesterol levels to be important. The results are reported in Table 3. First-generation modification to either the barley or the yeast to reduce costs was not valued by respondents, and they would require a price discount to be induced to purchase a beer with these characteristics. For both effects, the older the respondent, the less marked was their concern. For those who did not view cholesterol as important, genetic modification of yeast to generate health benefits was seen as neither positive nor negative: the coefficient is not significantly different from zero. However, those who did see cholesterol as an issue placed a positive weight on the health benefits, and would be prepared to pay more for this product. This was a product-specific effect—this group did not, in general, hold pro-GM views, as they did not hold a preference for first-generation GM changes.

The assumption of independence of irrelevant alternatives was tested by dropping the "conventional" alternative from the model, and re-estimating the model over the restricted, two-option data set (Hausman & McFadden, 1984). The null hypothesis, of no systematic difference in the parameter values, could not be rejected at conventional levels of significance.

Table 4 reports the estimates with all insignificant variables removed.

Table 3. Parameter estimates from a conditional logit model.
LL value = -385.30 Choice sets = 410
  Coeff Std.Err Z p
Price -1.356 0.150 9.01 0.00
Barley -0.913 0.355 2.57 0.01
Barley * Age 0.017 0.008 2.21 0.03
Yeast(1) -1.810 0.562 3.22 0.00
Yeast(1) * Age 0.021 0.011 1.92 0.06
Yeast(1) * Chol 0.366 0.539 0.68 0.50
Yeast(2) -0.152 0.447 0.73 0.73
Yeast(2) * Age 0.006 0.009 0.67 0.50
Yeast(2) * Chol 0.877 0.412 2.13 0.03
Notes: Barley = 1 if includes first-generation barley, 0 otherwise
Yeast(1) = 1 if includes first-generation yeast, 0 otherwise
Yeast(2) = 1 if includes cholesterol enhancement yeast, 0 otherwise
Price = price of beer
Age = age of respondent
Chol = 1 if respondent viewed cholesterol level to be important, 0 otherwise

Table 4. Parameter estimates from a conditional logit model: significant variables only.
LL value=-385.84 Choice sets = 410
  Coeff Std.Err z p
Price -1.351 0.150 8.99 0.00
Barley -0.979 0.337 2.90 0.00
Barley * Age 0.019 0.007 2.55 0.01
Yeast(1) -1.499 0.438 3.42 0.00
Yeast(1) * Age 0.021 0.009 2.22 0.03
Yeast(2) * Chol 0.966 0.155 6.19 0.00

What is notable about these results is the relative similarities in the size of the coefficients on the first-generation GM variables (Barley and Yeast(1)) and the effects of age. This suggests that there may be no difference in attitudes with respect to the vehicle of the modification (plant or microorganism). This can be tested formally by restricting the parameters on these variables to be equal, as in Table 5. This restriction is accepted on the basis of a LL test (test statistic of 4.54, compared to a critical value of 5.99), leading to a very parsimonious representation of preferences.

Table 5. Parameter estimates from a conditional logit model: equality of 1st generation effects.
LL value = -388.11 Choice sets = 410
  Coeff Std.Err z p
Price -1.286 0.150 8.83 0.00
Barley -1.150 0.283 4.04 0.00
Barley * Age 0.019 0.006 3.16 0.00
Yeast(1) -1.150 0.283 4.04 0.00
Yeast(1) * Age 0.019 0.006 3.16 0.00
Yeast(2) * Chol 1.09 0.144 7.56 0.00

The implication of this specification is that the presence of both a cost-reducing GM barley and yeast had twice the impact on the consumers' valuation of the beer, as compared with either one on its own. However, it is possible that the respondents did not make a distinction between the degree of genetic modification; once they identified any level of first-generation genetic modification, it was sufficient to induce an adverse effect. Of the sample of 1,230 beers presented, 615 involved one first-generation GM process, while 82 had two (i.e., both cost-reducing barley and cost-reducing yeast). To test whether there was any marginal reduction in utility associated with the second GM process, two new variables were created. The first (One) took a value of 1 if there was one first-generation GM process involved in the production of the beer, and 0 otherwise. The second variable (Two) took a value of 1 if two first-generation GM processes were involved in the production of the beer, and 0 otherwise. These variables were then used to replace the individual barley and yeast variables used before. Neither the variable Two nor Two*Age were significant, implying that the presence of a second first-generation GM process did not alter the respondents' valuation of the product. The final form of the estimated model is reported in Table 6.

Although these results indicate only the signs and significance of effects, they can be given monetary values by identifying the partworths associated with changes in attribute levels. These are defined by the negative ratio of attribute to price coefficient. Table 7 reports partworths for first-generation GM and functional GM for those concerned about cholesterol levels.

Table 6. Parameter estimates from a conditional logit model: final specification.
LL value = -385.92 Choice sets = 410
  Coeff Std.Err Z p
Price -1.386 0.155 8.92 0.00
One -1.440 0.330 4.36 0.00
One * Age 0.022 0.007 3.12 0.00
Yeast(2) * Chol 1.15 0.148 7.78 0.00
Note: One = 1 if one first-generation process is involved (either barley or yeast), 0 otherwise.

Table 7. Partworths associated with genetic modification and beer.
  Partworth ($) p
1st generation GM(age=20) -0.72 0.00
1st generation GM (age=40) -0.40 0.00
Functional GM 0.83 0.00

Thus, the presence of first-generation GM of either form would require a discount of A$0.72 for the 20-year-old respondent to be left indifferent as compared to the conventional beer; this declines to A$0.40 for 40-year-old respondents. However, those who viewed cholesterol as a significant issue for themselves would be prepared to pay a premium of A$0.83 to purchase a bottle of beer that had the described medical implications. Those who did not see cholesterol as an issue were indifferent to the presence of this modification; it neither increased nor reduced their perception of the beer.

Conclusions

Although the sample size was small, the repeated nature of the experiments means that the preferences of this set of respondents can be identified with high precision. The results reveal a diversified set of preferences towards genetic modification in foods. There was a set of respondents (30% in this sample) who were not prepared to select a beer having any GM component in its production for any of the price or health advantages offered in this experiment. There was a set of respondents who required some price discount to be induced to purchase a beer that had some first-generation GM involved in its production. However, this effect appeared not to be cumulative—the presence of a single first-generation GM process was sufficient to generate a reduction in utility, but subsequent additional GM processes did not further extend this. There was a third set who were prepared to pay a premium to access a product having medicinal benefits. It was particularly reassuring that this subset corresponded to those who revealed a concern about cholesterol in other areas of the survey; this gave some support to the validity of the survey instrument. However, the expected differentiation between GM plants and GM microorganisms was not present in this sample; a similar level of concern was found for both.

An unresolved question is whether concerns about first-generation GM products would be moderated by market exposure. If this were the case, there would be significant benefits for those striving for market acceptance to develop products with direct nutritional or health benefits to consumers, as opposed to some unspecified price advantage. A second issue is the inevitably conditional nature of the preferences that the choice modeling framework reveals. The assumption made here is that the discount reflects some disutility associated with the process. However, as noted by a reviewer, it may be that respondents were expressing a view that any cost savings associated with the use of first-generation GM (which was how the technology was motivated in the survey) should be passed on to consumers. Hence, even if they were indifferent to the product, they were expressing a preference for market consequences of its use based on some notion of equity. The current survey was not designed to tease out these possibilities, but it does show the potential complexity of consumer responses to the introduction of these technologies.

Appendix: The Survey Instrument

The survey consists of four sections over four pages. Click here for a PDF version of the survey instrument.

References

Adamowicz, W., Boxall, P., Williams, M., & Louviere, J.J. (1998) Stated preference approaches for measuring passive use values: Choice experiments and contingent valuation. American Journal of Agricultural Economics, 80, 64-75.

Baker, G.A. and Burnham, T.A. (2001) Consumer response to genetically modified foods: Market segment analysis and implications for producers and policy makers. Journal of Agricultural and Resource Economics, 26(2), 387-403.

Bennett, J. and Blamet, R. (Eds.) (2001). The choice modelling approach to environmental evaluation. Northampton, MA: Edward Elgar.

Burton, M., Rigby, D., Young, T., & James, S. (2001). Consumer attitudes to genetically modified organisms in food in the UK. European Review of Agricultural Economics, 28(4), 479-498.

Donaghy, P., Rolfe, J., & Bennett, J. (2002, February). Disaggregating consumer demands for organic and genetically modified foods using the choice modelling technique. Paper presented at AARES 2002: the 46th Annual Conference of the Australian Agricultural and Resource Economics Society, Canberra, Australia.

Gamble, J., Muggleston, S., Hedderley, D., Parminter, T., & Richardson-Harman, N. (2000). Genetic engineering: The public's point of view (report to stakeholders). The Horticulture and Food Research Institute of New Zealand Ltd. Available on the World Wide Web: http://www.hort.cri.nz/media/gm_report.pdf.

Greene, W.H. (1997). Econometric analysis (3rd ed.). Englewood Cliffs, NJ: Prentice Hall.

Hausman, J. and McFadden, D. (1984). Specification tests in econometrics. Econometrica, 52, 1219-1240.

Louviere, J.J., Hensher, D.A., & Swait, J.D. (2000). Stated choice methods: Analysis and applications. Cambridge, UK: Cambridge University Press.

McFadden, D. (1973). Conditional logit analysis of qualitative choice behaviour. In P. Zarembka (Ed.), Frontiers in Econometrics. New York: Academic Press.

Morrison, M.D., Blamey, R.K., Bennett, J.W., & Louviere, J.J. (1996). A comparison of stated preference techniques for estimating environmental values. Choice Modelling Research Reports. The University of New South Wales.

Owen, K., Louviere, J., & Clark, J. (2002, February). Consumer concern and acceptance of GM foods. Paper presented at AARES 2002: the 46th Annual Conference of the Australian Agricultural and Resource Economics Society, Canberra, Australia.

Schmidt, D.B. (2000). Consumer response to functional foods in the 21st century. AgBioForum, 3(1), 14-19. Available on the World Wide Web: http://www.agbioforum.org.


Suggested citation: Burton, M., & Pearse, D. (2002). Consumer attitudes towards genetic modification, functional foods, and microorganisms: A Choice modeling experiment for beer. AgBioForum, 5(2), 51-58. Available on the World Wide Web: http://www.agbioforum.org.
© 2002 AgBioForum | Design and support provided by Express Academic Services | Contact ABF: editor@agbioforum.org