Sorting biotechnology applications: Results of multidimensional scaling (MDS) and cluster analysis

Abstract

When people respond to questionnaires, they may construct preferences using various sources of information available, even questions within the questionnaire. An experimental approach with a representative sample of the Swiss population was applied to investigate how stable attitudes towards biotechnology are and which criteria people use spontaneously to evaluate and categorise biotechnology applications providing as little information as possible. A free card-sorting task using risk as a criterion versus not imposing a criterion was applied. Data were analysed using multidimensional scaling (MDS) and were represented in a cognitive map. Results of our experimental manipulation suggest that people’s preferences for biotechnology applications are relatively stable. Different sorting instructions did not result in different cognitive maps. Results suggest, therefore, that participants’ evaluations are not strongly influenced by criteria used in a questionnaire. The descriptions of the applications seem to be more crucial.

Keywords

card-sorting technique GM applications mental representations multidimensional scaling

1. Introduction

The implementation of new technologies such as biotechnology often depends on public judgement and acceptance. Research often concentrates on evaluating attitudes towards these technologies and factors influencing acceptance. In the field of biotechnology, many studies have been conducted to investigate either factors determining public acceptance or the perception of different applications. There is substantial evidence within different disciplines that public risk and benefit perceptions appear to drive beliefs about acceptability of biotechnology (Frewer et al., 2004; Verdurme and Viaene, 2003) along with ethical and moral considerations (Biel and Nilsson, 2005; Knight, 2007). Furthermore, researchers have shown that consumer acceptance of biotechnology varies among different applications (e.g. Frewer, Howard, Hedderley et al., 1997; Magnusson and Hursti Koivisto, 2002). If public concern is defined by the nature of application then perhaps the psychological constructs that underlie this concern are multidimensional, rather than defined by a single dimension such as perceived risk (Frewer et al., 1997). To predict public attitudes towards a specific application of biotechnology, it is necessary to take due account of these constructs and their importance for the public (Frewer, Howard, Hedderley et al., 1997). Most studies evaluating factors influencing the perception of biotechnology implemented structured questionnaires (e.g. Frewer et al., 1995; Gaskell et al., 1999). However, it has been shown that people construct preferences when asked to respond to survey questions and rely on both their knowledge and on other information available to them such as the questions themselves (Slovic, 1995). This may influence participants’ responses, and it is, therefore, a challenge for social science research to reduce the impact of potentially influential information or wording of questions on people’s responses. Providing the framework within which people have to make their evaluations provides only a limited view of public attitudes and the major factors that people use to make their evaluations. The advantage, however, is that the responses obtained directly translate into quantitative terms and require no further subjective interpretation (van Kleef et al., 2005).

In contrast, in an unstructured interview, the questions asked are not necessarily presented with the same wording to each participant, and participants can freely respond in their own words. This method requires further subjective interpretations, and the personal view of the researcher may influence data interpretation (van Kleef et al., 2005). Therefore research techniques offering further advantages should be implemented, meaning that participants’ responses should easily translate into quantitative terms without any subjective interpretation. It would also be advantageous not to ask questions that lead respondents in a specific direction. Examples of such research techniques are, for example, conjoint tasks and similarity ratings. In the area of biotechnology, conjoint analysis was used to explore consumer preferences for genetically engineered food products that provides information at the product characteristic level (Baker and Burnham, 2002; Frewer, Howard, Hedderley et al., 1997).

Similarity ratings allow the investigation of mental representations. They have been applied in only a few biotechnology studies (Frewer, Howard, Hedderley et al., 1997; Siegrist and Bühlmann, 1999). Frewer, Howard, Hedderley et al. (1997) found that specific applications were more differentiated in perceptual terms than general applications. Johnson and Tversky (1984) investigated laypeople’s perception of various hazards and suggested that the task used for data collection and the statistical model used to analyse the data influence the observed results. Similarity judgements did not correlate highly with rating scale evaluations. Similarity judgements are a more direct measure of the proximity between risks than rating scales. Siegrist and Bühlmann (1999) found that two dimensions were relevant for the perception of biotechnology applications: the nature of the applications and the organisms involved. Furthermore, property fitting revealed ethics, perceived risks and benefits, and acceptance to be important for the interpretation of the mental representation.

Our study investigated how stable attitudes towards biotechnology are and whether results of questionnaire studies can be reproduced using a completely different methodology without predetermined criteria. Furthermore, our study reveals which criteria people use spontaneously to evaluate and categorise biotechnology applications. We implemented an experimental design on a random sample of Swiss citizens to examine how laypeople categorise various biotechnology applications. Hence, the free sorting task using risk as a criterion versus not imposing any sorting criteria was applied. This should give insight into whether preferences regarding biotechnology are constructed due to predetermined criteria. We were also interested in testing whether it is possible to represent the data in a low-dimensional, meaningful cognitive map and which criteria people use to classify biotechnology applications.

2. Methods

Participants

‘Face-to-face’ interviews for the present study were conducted in 2008 in a Swiss community where nearby field trials with genetically modified (GM) wheat plants were carried out. An introductory letter containing general information about an interview regarding attitudes towards the field trials was sent to randomly chosen people taken from the telephone book. Soon afterwards, an interviewer contacted the selected households to arrange an interview date. Participants had to be 18 years or older, and they had to be able to speak and understand German. The response rate was 28% (n = 632).

Fifty-five percent (n = 347) of the participants were male, and 45% (n = 285) of the participants were female. The mean age was 53 years (SD = 16). Twenty-two percent of the participants were aged between 18 and 39 years, 49% between 40 and 64 years, 25% between 65 and 79 years, and 3% of the participants were over 80 years. Compared with the census data of Switzerland (BFS, 2009), men were over-represented, and the age and level of education were higher than the Swiss average.

At the beginning of each interview, the interviewer briefly introduced the topic and asked some general questions. Afterwards, the data collection began, and participants had to conduct a card-sorting task and fill in a questionnaire. Participants were not rewarded for participating in this study.

Card-sorting task

Participants were confronted with 29 cards (8 × 6 cm), each with a different biotechnology application written on it. Special attention was paid to include medical, nutritional, agricultural, and industrial applications of biotechnology as well as micro-organisms, bacteria, human stem cells, plants and animals (see Table 1). Each card had an identification code on the top left corner to simplify recording, which was not related to any order or relatedness of the cards.

Table 1.

29 card stimuli with their abbreviations.

Application	Abbreviation
Genetically modified bacteria that produce insulin for diabetics.*	Bacterial insulin
Isolation of human embryonic cells to produce organs for transplantations.	Embryonic cells producing organs
Genetically modified micro-organisms that produce vaccines against infectious diseases.*	Micro-organisms producing vaccines
Genetically modified pigs, which can produce human organs for transplantations.*	Pigs producing human organs
Gene diagnoses to detect diseases such as HIV or cancer.	Gene diagnosis (cancer)
Application of gene diagnostics to discover foetal genetic defects such as Down syndrome.	Gene diagnosis (Down syndrome)
Application of gene therapy for treating cancer.	Gene therapy (cancer)
Application of gene therapy to cure genetic diseases. Healthy genes are injected into an ill person.	Gene therapy (hereditary diseases)
Genetically modified rice that produces vitamin A.	Golden rice
Genetically modified micro-organisms that produce rennet for cheese production.*	Rennet
Genetically modified soy plants that produce more oleic acid and therefore have a higher content of unsaturated fatty acids.*	Oleic acid enhanced soy
Genetically modified oil seed rape, which lowers the level of cholesterol.	Cholesterol reducing rape
Genetically modified lettuce with an enlarged vitamin C content.	Vitamin C enhanced lettuce
Genetically modified cereals, which do not contain gluten. Therefore, bread and cakes made using their flour are suitable for people with allergies.*	Gluten free cereals
Genetically modified salmon that grow bigger and faster to produce more meat.*	Bigger salmon
Genetically modified tomatoes that do not get soft too quickly.	Tomatoes (texture)
Genetically modified cows, which produce milk similar to human breast milk.*	Cow milk for babies
Genetically modified cows that produce milk without lactose.*	Lactose free milk
Genetically modified tomatoes that have a more aromatic taste.*	Tomatoes (taste)
Genetically modified potatoes that contain more starch, therefore reducing the amount of fat needed to deep-fry.	Starch enhanced potatoes
Genetically modified corn that is resistant to pests.*	Pest resistant corn
Genetically modified wheat that is resistant to diseases.	Disease resistant wheat
Genetically modified apples that are resistant to apple scab.	Scab resistant apples
Genetically modified oilseed rape that is resistant to pests.	Pest resistant rape
Genetically modified agricultural crops that are resistant to herbicides.	Herbicide resistant crops
Genetically modified wheat that is resistant to fungi.	Fungi resistant wheat
Genetically modified trees that grow faster to produce paper and wood.	Trees
Genetically modified enzymes that are in washing powder for more effective cleaning.*	Washing powder
Genetically modified plants that produce more fat and oils for bio fuel.	Bio diesel

Note. * indicates whether the statement was evaluated in the written questionnaire.

The high number of applications did not allow pair-wise similarity ratings, as participants would have had to compare 406 pairs of applications. Therefore, the free card-sorting technique was applied. The data were analysed using MDS, cluster analysis and property fitting. The basic idea behind the sorting techniques is simply to ask respondents to sort different objects into groups. The advantages of this free-sorting method are that it is not a taxing technique and it corresponds closely to natural and mental activities (Coxon, 1999).

Participants were randomly assigned into one of two groups. One group (155 women, 177 men) was instructed to sort the cards into groups without specifying any criteria, and the other group (130 women, 170 men) was asked to separate the cards into groups using risk as a criterion. At the beginning, participants were given a brief explanation of their sorting task and had some time to familiarise themselves with the applications. Participants were asked to sort the applications into between two and 28 categories. Each application could be assigned in only one category, and all of the applications had to be sorted. No time limit was imposed. At the end, a questionnaire was administrated to obtain further information about the participants.

Questionnaire

Twelve applications from the 29 applications of the sorting task were selected. These applications are designated with a * in Table 1. Special attention was taken to ensure that the applications encompassed plant, animal and micro-organisms along with medical, nutritional, agricultural and industrial aspects. Participants assessed the benefits and risks of these applications on a six-point scale (i.e. 1 = no risk/ no benefits and 6 = high risk/ high benefit). At the end of the questionnaire, sociodemographic characteristics, such as gender, age and level of education, were recorded.

The questionnaire and the applications on the cards were carefully worded in order to minimise academic language. The whole interview was pre-tested.

3. Results

Participants who were free to use any criterion formed 4.5 (SD = 1.7) categories. Participants who were asked using risk as a criterion formed on average 3.8 (SD = 1.6) categories. Significantly more categories were formed when participants were free to group the cards t(644) = 4.93, p < .001.

Different models can be used to analyse sorting data. The basic divergence among these models is between spatial (MDS) and discrete (cluster analysis) models. First, we analysed our data applying MDS. Therefore, a 29 × 29 similarity matrix was derived from each card-sorting task. Entries in these matrices reflect the number of participants who assigned the two applications to the same category.

For verification of differences between the two experimental groups, the data were analysed using the INDSCAL (individual differences scaling) method in SPSS 17 (available from SPSS, Inc.). This model exploits differences amongst a set of individual task matrices to achieve a unique orientation of the coordinate axes (Kruskal and Wish, 1978). The model links the similarities to the distances of the space via metric assumptions. In addition, the INDSCAL model makes a strong assumption about the communality of the space for the groups. The model presumes they all can be derived by differential stretching and shrinking the axes of a common space. In other words, it is possible that the groups differ in the number of dimensions needed for describing the data. The stress-I values for one to four dimensions (0.24, 0.12, 0.10, 0.09) were plotted in a graph. A decision was made to stop the analysis at four dimensions, because there was no substantial reduction in stress values between the third and fourth dimension. The stress-I values and the number of dimensions were plotted in a graph and showed a clear ‘elbow’ between the first and second dimension. The absolute stress-I value, as well as the interpretability of the solution, suggested the selection of the two-dimensional solution as additional dimensions do not result in a substantially better-fitting model (Kruskal and Wish, 1978). The dimension weights for the first dimension were 0.526 (task 1), 0.519 (task 2) and 0.446 (task 1), 0.445 (task 2) for the second dimension. No differences in the mental representation of the applications were found between the two groups. Results suggested that a two-dimensional solution describes the data very well. The experimental manipulation did not influence participants’ mental representation of biotechnology applications.

Owing to this finding, a similarity matrix of the full sample was computed and analysed utilising the ordinal PROXSCAL (multidimensional scaling of proximity data) method to find a least squares representation of the objects in a low-dimensional space (Coxon, 1999). Stress-I values for one- to four-dimensional solutions were 0.24, 0.09, 0.08, 0.07, respectively. This analysis supported the INDSCAL results that a two-dimensional solution explains the data well. Results of the MDS analysis are shown in Figure 1.

Figure 1.

Two-dimensional solution and results of the cluster analysis (dotted circles) of the 29 biotechnology applications 189x115mm (300 x 300 DPI).

For stability verification, we also performed a hierarchical cluster analysis using the method of squared Euclidian distances to place greater weight on objects that are further apart, and the average linkage between groups was applied. This is considered to be a fairly robust method (Coxon, 1999). The results of the cluster analysis are also shown in Figure 1 as dotted circles.

Participants grouped the applications in two large clusters with one encompassing all medical applications. There are two sub-clusters within the ‘medical’ cluster; one includes medical treatments and diagnoses of illnesses, and the other concerns stem cell research. The other big cluster encompasses all ‘non-medical’ applications and is subdivided into two big sub-clusters: one combines all plant and industrial related applications, which splits into two sub-clusters, and the other encompasses crop plant-related applications. Within that cluster, there are two sub-clusters: one comprises second-generation crop plants, and the other encompasses plant-related applications that have certain benefits for nutrition and plants that are tolerant against herbicides. The second sub-cluster in the ‘plant/industrial’ cluster involves industrial applications. Within this cluster are two sub-clusters: one includes applications concerning micro-organisms and enzymes. The second combines the industrial applications of plants as well as plants that show more positive characteristics. The other ‘non-medical’ sub-cluster includes all applications involving animals. One of the ‘animal’ sub-clusters includes two applications involving cows, which then form a cluster together with the ‘salmon’. These three applications are connected with the ‘pig’ application.

So far we have provided a regional interpretation of the MDS solution. We were also interested, however, in whether the two dimensions of the solution can be meaningfully interpreted. Property fitting is one method that can be used to facilitate a dimensional interpretation of an MDS solution (Kruskal and Wish, 1978). Multiple regression analysis was used to identify variables that are correlated with the dimensional weights of the applications. Participants indicated perceived risks and benefits for 12 applications, as assessments of all 29 applications would have been too time-consuming. The mean risks and benefits were then used as dependent variables and dimensional weights as predictor variables. Standardised results show that the first dimension can be clearly explained with the property fitting but not the second dimension (benefit: β_dimension1 = .57, β_dimension2 = .64, R² = .82, p < .01, risk: β_dimension1 = .02, β_dimension2 = −.64, R² = .27, p = .10).

4. Discussion

Most people may not have preferences for specific biotechnology applications, but people construct preferences when asked to respond to questions on this topic (Slovic, 1995). Those participating in surveys may construct their preferences when they answer questions and can use whatever information they have available, even the questions in the questionnaire. Asking about possible risks of biotechnology applications, people may infer that biotechnology involves risk, because the question would not be asked otherwise. In other words, people may be primed with information provided in questionnaires or interviews. It is unclear how malleable people’s preferences towards biotechnology applications are, and how strongly preferences are influenced by the contextual information provided in a questionnaire. In order to examine this question, we used an experimental approach. One group was asked to sort the cards into groups without specifying any criteria, and the other group was asked to sort the cards into groups using risk as a criterion. We did not find any differences in people’s mental representations of biotechnology applications. Regardless the criterion provided, people’s cognitive map did not change. This gives evidence that preferences for biotechnology applications are not constructed through priming of risks. Nonetheless, it cannot be completely excluded that other contextual factors may have influenced people’s perceptions. They reflect rather stable preferences. Therefore, results gained from questionnaire or interview studies about biotechnology mirror people’s perceptions well (e.g. Bredahl, 1999; Frewer et al., 1996). Our study verifies previous studies utilising a completely different research method.

The analysis of the combined matrix for both tasks revealed a two-dimensional map, with the first dimension clearly relating to perceived benefits. Genetically modified products associated with tangible benefits for the consumer are more likely to be accepted than products without perceived benefits (Hossain et al., 2003). When the product is of a medical nature, the chances of it being accepted by the public are greater because people perceive more benefits and fewer risks from medical applications (e.g. Connor and Siegrist, 2010; Frewer, Howard, Hedderley et al., 1997). All applications that are obviously of a medical nature cluster together and are perceived to be more beneficial than the other applications. This result is in line with previous studies (Magnusson and Hursti Koivisto, 2002). However, other studies found that people disapprove of the use of human genetic testing and may have moral objections towards the manipulation of human DNA (Frewer, Howard, Hedderley et al., 1997; Gaskell, 1998).

The second dimension cannot clearly be associated with one specific psychological factor. One possible interpretation for this dimension could be ethical or moral objections; another could be the organism involved, which does not exclude the ethical factor and is consistent with other studies that determined ethics and morality to be an important factor (Knight, 2007; Lassen and Jamison, 2006). Overall, results suggest that people considered the organisms involved in the application rather than just the type of application.

Results of the present study suggest that the card-sorting task provided similar insights to the studies in which researchers explicitly stated the dimensions participants should use for their evaluation. In the field of biotechnology, many studies have been published examining people’s perception. As a result, researchers may have a good idea which dimensions lay people use to evaluate biotechnology applications. The sorting task might be a very valuable tool for examining public perception of new technologies or hazards for which the dimensions that lay people use for their evaluations are unknown. By utilizing the method of sorting or similarity ratings, one can avoid the mistake of not including dimensions in a questionnaire that may be important to respondents. This is the case because, in sorting tasks, participants are free to use any dimension they perceive as important for completing their task. In other risk perception studies, similarity ratings were used rather than sorting tasks. These studies found that similarity ratings may be useful in predicting people’s responses to new risk or to new evidence about risk (Johnson and Tversky, 1984). Similarity ratings are only possible, however, when there is a limited number of stimuli. The advantage of the sorting task combined with MDS, cluster analysis and property fitting is the fact that it can also be used for a large number of stimuli. Further, the advantages of the card-sorting method are that it is not a taxing technique and it corresponds closely to natural and mental activities (Rugg and McGeorge, 1997). Card sorting is also applicable to non-literate cultures and small children through the presentation of pictures. Therefore, we encourage the implementation of card-sorting tasks for future studies in the field of risk perception.

Footnotes

Funding

This study was funded by the Swiss National Science Foundation. The National Research Programme NRP 59: Benefits and Risks of the Deliberate Release of Genetically Modified Plants.

Author biographies

Melanie Connor is a doctoral student at the Institute for Environmental Decisions in the research group on Consumer Behavior at ETH Zurich, Switzerland. Her research interests lie in people’s perception of new technologies, risk perception and consumer behaviour.

Michael Siegrist is Professor at ETH Zurich, Switzerland, and is chair of the research group on Consumer Behavior. His research interests lie in risk perception, risk communication and acceptance of new technologies. A specific topic is consumer behaviour with regard to food.

References

Baker

Burnham

(2002) The market for genetically modified foods: Consumer characteristics and policy implications. International Food and Agribusiness Management Review 4: 351–360.

BFS. (2009) Statistik des jährlichen Bevölkerungsstandes (ESPOP) und der natürlichen Bevölkerungsbewegung (BEVNAT) 2008. Provisorische Ergebnisse (Swiss national census data 2008). Neuchâtel: Bundesamt für Statistik.

Biel

Nilsson

(2005) Religious values and environmental concern: Harmony and detachment. Social Science Quarterly 86(1): 178–191.

Bredahl

(1999) Consumers’ cognitions with regard to genetically modified foods: Results of a qualitative study in four countries. Appetite 33(3): 343–360.

Connor

Siegrist

(2010) Factors influencing people’s acceptance of gene technology: The role of knowledge, health expectations, naturalness, and social trust. Science Communication 32(4): 514–538.

Coxon

(1999) Sorting Data: Collection and Analysis (Vol. 07-127) Thousand Oaks, CA: Sage.

Frewer

Howard

Shepherd

(1995) Genetic engineering and food: What determines consumer acceptance? British Food Journal 97(8): 31–36.

Frewer

Howard

Shepherd

(1996) The influence of realistic product exposure on attitudes towards genetic engineering of food. Food Quality and Preference 7(1): 61–67.

Frewer

Howard

Shepherd

(1997) Public concerns in the United Kingdom about general and specific applications of genetic engineering: Risk, benefit, and ethics. Science, Technology, & Human Values 22(1): 98–124.

10.

Frewer

Howard

Hedderley

Shepherd

(1997) Consumer attitudes towards different food-processing technologies used in cheese production – The influence of consumer benefit. Food Quality and Preference 8(4): 271–280.

11.

Frewer

Lassen

Kettlitz

Scholderer

Beekman

Berdal

(2004) Societal aspects of genetically modified foods (proceedings paper). Food and Chemical Toxicology 42(7): 1181–1193.

12.

Gaskell

(1998) Public perceptions of biotechnology in 1996. In: Durant

Bauer

Gaskell

(ed.) Biotechnology in the public sphere. London: Science Museum, 189–213.

13.

Gaskell

Bauer

Durant

Allum

(1999) Worlds apart? The reception of genetically modified foods in Europe and the US. Science 285(5426): 384–387.

14.

Hossain

Onyango

Schilling

Hallman

Adelaja

(2003) Product attributes, consumer benefits and public approval of genetically modified foods. International Journal of Consumer Studies 27(5): 353–365.

15.

Johnson

Tversky

(1984) Representations of perceptions of risk. Journal of Experimental Psychology 113(1): 55–70.

16.

Knight

(2007) Intervening effects of knowledge, morality, trust, and benefits on support for animal and plant biotechnology applications. Risk Analysis 27(6): 1553–1563.

17.

Kruskal

Wish

(1978) Multidimensional Scaling. Beverly Hills: CA: Sage.

18.

Lassen

Jamison

(2006) Genetic technologies meet the public. Science, Technology, & Human Values 31(1): 8–28.

19.

Magnusson

Hursti Koivisto

U-K

(2002) Consumer attitudes towards genetically modified foods. Appetite 39: 9–24.

20.

Rugg

McGeorge

(1997) The sorting techniques: A tutorial paper on card sorts, picture sorts and item sorts. Expert Systems 14(2): 80–93.

21.

Siegrist

Bühlmann

(1999) Die Wahrnehmung verschiedener gentechnischer Anwendungen: Ergebnisse einer MDS-Analyse (Perceptions of different applications of gene technology: Results of an MDS-analyses) Zeitschrift für Sozialpsychologie 30(1): 32–39.

22.

Slovic

(1995) The construction of preference. American Psychologist 50: 364–371.

23.

van Kleef

van Trijp

HCM

Luning

(2005) Consumer research in the early stages of new product development: A critical review of methods and techniques. Food Quality and Preference 16(3): 181–201.

24.

Verdurme

Viaene

(2003) Consumer beliefs and attitude towards genetically modified food: Basis for segmentation and implications for communication. Agribusiness 19(1): 91–113.