Monitoring consumer confidence in food safety: An ...

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/238368739

Monitoring consumer confidence in food safety: An exploratory study Article in British Food Journal · October 2004 DOI: 10.1108/00070700410561423

CITATIONS

READS

83

555

6 authors, including: Janneke de Jonge

Lynn Jayne Frewer

Belastingdienst

Newcastle University

39 PUBLICATIONS 702 CITATIONS

310 PUBLICATIONS 13,951 CITATIONS

SEE PROFILE

SEE PROFILE

Hans C. M. van Trijp

Reint Jan Renes

Wageningen University & Research

Wageningen University & Research

345 PUBLICATIONS 6,809 CITATIONS

19 PUBLICATIONS 333 CITATIONS

SEE PROFILE

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Thesis View project

Consumer Research on Novel and Safe Food View project

All content following this page was uploaded by Lynn Jayne Frewer on 08 June 2015. The user has requested enhancement of the downloaded file.

Risk Analysis, Vol. 30, No. 1, 2010

DOI: 10.1111/j.1539-6924.2009.01320.x

Consumer Confidence in the Safety of Food and Newspaper Coverage of Food Safety Issues: A Longitudinal Perspective Janneke de Jonge,1,∗ Hans Van Trijp,1 Reint Jan Renes,2 and Lynn J. Frewer1

This study develops a longitudinal perspective on consumer confidence in the safety of food to explore if, how, and why consumer confidence changes over time. In the first study, a theory-based monitoring instrument for consumer confidence in the safety of food was developed and validated. The monitoring instrument assesses consumer confidence together with its determinants. Model and measurement invariance were validated rigorously before developments in consumer confidence in the safety of food and its determinants were investigated over time. The results from the longitudinal analysis show that across four waves of annual data collection (2003–2006), the framework was stable and that the relative importance of the determinants of confidence was, generally, constant over time. Some changes were observed regarding the mean ratings on the latent constructs. The second study explored how newspaper coverage of food safety related issues affects consumer confidence in the safety of food through subjective consumer recall of food safety incidents. The results show that the newspaper coverage on food safety issues is positively associated with consumer recall of food safety incidents, both in terms of intensity and recency of media coverage. KEY WORDS: Confidence; food safety; incidents; media attention; monitoring; newspaper coverage; recall; risk; trust

fidence in the safety of food in general. One of the priorities of food safety authorities is to generate consumer confidence in the safety of food. Monitoring consumer confidence over time is important to evaluate the impact of actions of food safety institutions directed at improving the risk analysis process and protecting the public from food risks. This study develops and validates such a monitoring instrument. The monitor is based on the framework of consumer confidence in the safety of food developed by de Jonge et al.(4) Within this framework, consumer trust in various actors in the food chain with responsibility for consumer protection, the perceived safety of a range of product groups, and consumer recall of food safety incidents have been identified as key factors that drive general confidence. There are two reasons why effective monitoring of consumer perceptions is not a trivial task. First, developments over time have often been investigated in

1. INTRODUCTION Despite the fact that, in developed countries, food safety standards and food quality performance are reported to be higher than previously,(1) various food safety incidents have occurred over the past few decades.(2,3) In addition to affecting food safety perceptions related to specific product groups, the accumulation of incidents, no matter how different in character and in terms of consequences for public health, may also put pressure on consumer con1

Wageningen University, Marketing and Consumer Behaviour Group, Wageningen, The Netherlands. 2 Wageningen University, Communication Science, Wageningen, The Netherlands. ∗ Address correspondence to Janneke de Jonge, Wageningen University, Marketing and Consumer Behaviour Group, Hollandseweg 1, 6706 KN, Wageningen, The Netherlands; [email protected].

125

C 2009 Society for Risk Analysis 0272-4332/10/0100-0125$22.00/1

126 previous studies by comparing consumer responses to individual questions between subsequent surveys, rather than the interrelationships between the measures.(3,5−7) Although this may provide (preliminary) insight into directions of change in perceptions, potential sources of these changes are difficult to identify. Therefore, to understand why changes in consumer confidence occur, this study investigates changes in consumer confidence in direct relationship to its determinants. Second, from a methodological point of view, comparisons over time can only be interpreted unambiguously if model and measurement validity are satisfied. Few previous studies have assessed measurement equivalence,(3,5,6,8) which means that there is more uncertainty regarding whether the identified changes over time are real changes (i.e., substantive changes in content), or a result of measurement error. It has been argued that the mass media can play an important role in building and undermining consumer confidence in the safety of foods,(9) particularly because consumers have limited ability to assess food safety prior to consumption. In their assessment and evaluation of food safety, consumers, therefore, rely heavily on information provided by others. Media coverage of food safety issues has primarily been studied in relation to specific food incidents and food products. This research confirms that the occurrence of incidents and media coverage of these incidents are likely to influence consumer perceptions about the safety of specific product groups or types of food.(9,10) However, this line of research has not addressed how daily media reporting about the totality of food safety related issues may accumulate to affect general consumer confidence in the safety of food. This study addresses this question by monitoring actual newspaper coverage on the totality of food safety issues in parallel to consumer recall of food safety incidents. 1.1. Research Aim The first aim of this study is to develop and empirically validate a monitoring instrument for consumer confidence in the safety of food, which is theoretically founded and methodologically sound. Such a monitoring instrument will allow for formal comparisons of interrelationships between constructs and construct means over time (Study 1). The second aim of this study is to examine to what extent newspaper coverage of food safety related issues affects consumer recall of food safety incidents (Study 2).

de Jonge et al. 2. STUDY 1. MONITORING CONSUMER CONFIDENCE OVER TIME 2.1. A Framework for Consumer Confidence in the Safety of Food Confidence judgments are relevant for many issues in life,(11) such as personal abilities,(12) future economic developments,(13) and the safety of food.(4) General confidence has been defined as “the belief that most future events will occur as expected.”(11) In the context of food safety, confidence represents the implicit belief that the consumption of food products will not result in adverse health effects, as this is what the average consumer would normally expect to happen. That is, confidence is based on familiarity,(11) and develops from the accumulation of positive experiences. The concept of general consumer confidence in the safety of food has been conceptualized along two distinct, but correlated, dimensions, that is, optimism and pessimism.(14) The framework of consumer confidence in the safety of food developed by de Jonge et al.(4) identifies four determinants of general consumer confidence in the safety of food: trust in regulators and actors in the food chain, the perceived safety of product groups, consumer recall of food safety incidents, and individual differences. Consumer trust in regulators, producers, and distributors responsible for the management of hazards is considered to be an important driver of general consumer confidence in the safety of food(15,16) because trust enables consumers to compensate for the lack of knowledge about the safety of food they eat.(17−19) Conferring trust to a particular entity allows people to take things for granted,(20) and to behave according to habits or routines.(21) Perceptions of expertise, honesty, and benevolence are regarded as underlying dimensions of trust.(22) The second determinant of general consumer confidence in the safety of food is the perceived safety of product groups.(3) Because food scares are associated with potential risks related to one or several particular product groups, the perceived safety of product groups is expected to influence general consumer confidence in the safety of food. The occurrence of food safety incidents has often resulted in increased consumer perceptions of risk.(23) Therefore, consumers who recall food safety incidents can be expected to be less confident about the safety of food in general, compared to consumers to whom food safety incidents are not salient.

Consumer Confidence in the Safety of Food and Newspaper Coverage of Food Safety Issues Finally, individual difference variables are included in the framework as a determinant of general consumer confidence in the safety of food because individuals tend to differ regarding the extent to which they are concerned about food-related hazards,(24,25) and food safety in general.(26,27) 2.2. Monitoring Changes in Consumer Confidence Monitoring on the basis of longitudinal data has several advantages over cross-sectional data. An advantage is that longitudinal data provide “base level” measurements against which changes in consumer perceptions can be captured. Two types of changes over time can be distinguished. The first lies in changes in the strength of the relationships between the constructs of the model. Second, changes in the mean ratings on the constructs of the model can be assessed. Monitoring on the basis of longitudinal data introduces a number of methodological challenges that are often overlooked. Research in, among others, psychometrics,(28) organizational research,(29) and marketing(30) has pointed to the fact that comparisons over time can only meaningfully be made if the data conform to a number of measurement invariance criteria. That is, comparisons over time can only be justified methodologically if there is sufficient evidence that construct measures have the same content and meaning across different measurement occasions. Structured approaches(28,30,31) that allow for such formal comparisons have been developed within the context of multigroup structural equation modeling (e.g., LISREL) to systematically assess the measurement properties of the model and to statistically explore changes over time. In this study, such a structured approach(32,33) is applied in the context of the monitor on consumer confidence in the safety of food. 2.3. Method 2.3.1. Sample Consumer perceptions about the safety of food have been assessed in four annual surveys. Data collection took place during a three-week period in November and December in all four years. Data were collected by a professional market research agency (GfK Panelservices Benelux B.V.), and quota sampling was performed on the basis of gender, age, education level, household size, and area of resi-

127

dence. Administering the survey took place through the Internet. In the Netherlands, Internet access from home is common (83%). Even the segment with the lowest Internet usage from home (lower educated persons) has an access rate of 73%.(34) The four surveys were conducted with different respondents. In total, 4,458 respondents were invited to take part in the research, and 2,504 respondents filled out the questionnaire (a response rate of 56.8%). The sociodemographic make-up of the samples regarding gender, age, education level, and household size were compared against national population statistics on these variables (Table I). Males were somewhat overrepresented, particularly in 2003. Across all fours years, older consumers (older than 50 years old) and single person households were underrepresented, and consumers with a high education level were overrepresented. Although in terms of these sociodemographics, the sample cannot be regarded as completely representative of the Dutch population, all samples cover a broad range of sociodemographic backgrounds. Respondents with more than four missing observations out of 45 items (6.9%), and respondents who had missing values on all items of a particular construct (0.4%, beyond respondents with >4 missing observations) were not included in the analysis. For the remaining respondents (92.7%), an estimation of the missing values was made using two-way imputation(35) for each sample individually. This resulted in 515 observations suitable for analysis in 2003, 616 in 2004, 577 in 2005, and 614 in 2006. 2.3.2. Materials Measures (see Appendix A) to assess the key constructs of general consumer confidence in the safety of food (optimism and pessimism), consumer trust in different actors in the food chain (the government, farmers, retailers, and the food industry), and the perceived safety of a range of different product groups were identical to those reported in de Jonge et al.(4) The constructs of perceived control and trait worry were not included in the current longitudinal analysis because data on these constructs were not available in 2003. Consumer recall of food safety incidents was included in the framework as a dummy variable, which indicated whether a consumer recalled an incident or not. When respondents responded that they recalled a food safety incident, they were, in addition, asked to indicate what incident they recalled. Each respondent could enter up

de Jonge et al.

128 Table I. Sample Characteristics (%)

Data Collection Period

2003 (n = 563) November 27– December 10




58.5

52.6

49.4

66.7

66.8 33.2

56.9 43.1

54.3 45.7

55.9 44.1

49.1 50.9

8.7 10.1 9.1 20.2 17.4 19.5 14.9

8.8 6.8 10.4 21.8 18.1 19.8 14.3

7.6 6.4 10.0 21.1 18.8 20.7 15.5

7.7 7.9 9.8 20.8 18.4 20.8 14.7

7.4 7.3 7.5 18.8 18.9 23.0 17.2

21.5 43.5 35.0

28.3 39.9 31.8

25.2 37.3 37.5

30.6 39.8 29.6

33.6 41.2 25.2

11.4 37.1 51.5

11.6 39.3 49.2

12.5 37.0 50.5

13.8 47.3 38.9

34.5 32.7 32.8

65.5 15.3 14.7 4.4

65.0 16.3 14.3 4.4

64.3 16.1 14.1 5.4

72.9 17.5 8.0 1.6

19.5 80.5

18.7 81.3

18.8 81.3

18.8 81.2

Response rate Gender Male Female Age 15–19 20–24 25–29 30–39 40–49 50–64 65+ Education level Low Average High Household size 1 person 2 persons 3 or more persons Number of children 0 1 2 3 or more Allergic Yes No a Source:

Population Statistics January 1, 2005a

Statistics Netherlands.

to three recalled incidents. Finally, information was collected regarding respondents’ gender, education level, age, the number of children in the respondent’s household, and household experience of food allergy or food intolerance. 2.3.3. Data Analysis The four waves of data were analyzed using multigroup structural equation modeling where each wave of data collection was considered as a separate group. By simultaneously estimating the model for the different measurement occasions (i.e., the four groups), it could be established whether the properties of the measurement model were stable over time, whether the mean scores on the constructs differed over time, and whether the relative importance of the determinants of confidence changed over time. LISREL 8.72 software was used for the estimation. Maximum likelihood estimation was employed using covariances as input for the analysis. To iden-

tify the model, one item per construct was defined as a marker item, indicated by an asterisk in Appendix A, with a factor loading of one, and an intercept of zero. Items with a moderate amount of variance were chosen as marker items.(36) Regarding the single-item measures covering sociodemographic information and recall of incidents, an assumption of no error was made. Estimations of the latent means are dependent upon which item is used as the marker item because the scale of the construct mean is set to be equivalent to that of the chosen marker item. As such, the absolute values of the latent means should not be considered as “true” scale values, and only be used for comparing means of the same construct over time. The analytic strategy involved three steps. First, the measurement models were evaluated and tested on measurement equivalence. The measurement model consisted of both dependent variables (optimism and pessimism) as well as their determinants (i.e., trust in actors, perceived safety of

Consumer Confidence in the Safety of Food and Newspaper Coverage of Food Safety Issues product groups, recall, sociodemographic variables). In the establishment of the measurement model, the relationships between the determinants and the two dependent variables (i.e., the structural parameters) were not estimated because testing group differences between the structural parameters was part of the second step of the analysis. The general approach as outlined by Steenkamp and Baumgartner(30) was followed, where, in a sequence of steps, more stringent constraints were cumulatively imposed on the parameters. Consecutively, the measurement model was tested for configural invariance, metric invariance, scalar invariance, factor covariance invariance, factor variance invariance, and error variance invariance. Second, differences over time in the structural parameters (i.e., regression coefficients) were investigated. Assessments of differences between the structural relationships over time were made by establishing the change in model fit of two competing models: one in which the structural relationships were freely estimated for each wave of data collection, and one where these relationships were constrained to be equal over time. In the third step, longitudinal changes in the mean ratings on the latent constructs were assessed. Model fit and comparison between subsequent models was assessed on the basis of χ 2 , the root mean square error of approximation (RMSEA), the comparative fit index (CFI), and the consistent Akaike information criterion (CAIC).(29) The RMSEA and CFI range from 0 to 1. An RMSEA value below 0.08 indicates a reasonable fit, and when RMSEA is below 0.05 the model is considered to closely fit the data.(37) For CFI, higher values indicate better model

129

fit. Typically, 0.95 ≤ CFI < 0.97 is considered to be an acceptable fit, and 0.97 ≤ CFI ≤ 1.00 is considered a good fit.(38) The χ 2 change of subsequent models should be insignificant. However, as the test is affected by sample size, for large samples relatively modest differences can become significant.(38) Therefore, the extent to which model improvement had been achieved was also evaluated with CAIC, which adjusts for model parsimony. Lower values of CAIC for alternative models are an indication of a better model fit. 2.4. Results Following Steenkamp and Baumgartner,(30) the measurement model was tested in a sequence of steps, where invariance constraints were cumulatively imposed on the parameters. In the least restricted measurement model (i.e., the configural invariance model), all parameters were estimated individually for each group, which resulted in a total number of 956 estimated parameters. The fit statistics of the increasingly constrained measurement models are shown in Table II. The configural invariance model showed a good fit to the data: χ 2 (4552) = 14935.25, p < 0.01; RMSEA = 0.065; CFI = 0.96; and CAIC = 24135.28. By stepwise imposing the different invariance constraints onto the model, it was investigated whether the model fit would remain the same, which would be an indication of measurement invariance. CFI and CAIC were used to evaluate changes in overall model fit of the different models, where CFI should remain the same, and CAIC should decrease, for subsequent models with more strict invariance constraints. Individual parameters that were not

Table II. Model Fit Statistics of Increasingly Constrained Measurement Model

Configural invariance Full metric invariance Full scalar invariance Final partial scalar invariancea Full covariance invariance Full variance invariance Final partial variance invarianceb Full error variance invariance Final partial error variance invariancec

χ2

df

RMSEA

CFI

CAIC

χ 2

df

p

14935.25 15066.28 15263.12 15184.81 15609.16 15744.12 15675.76 16196.14 16023.24

4,552 4,657 4,762 4,757 5,033 5,081 5,079 5,214 5,207

0.065 0.065 0.064 0.064 0.063 0.063 0.062 0.063 0.063

0.96 0.96 0.96 0.96 0.96 0.96 0.96 0.96 0.96

24135.28 23371.52 22649.28 22605.78 20663.06 20379.08 20329.23 19826.59 19635.26

131.03 196.84 118.53 424,35 134.96 66.60 520.38 347.48

105 105 100 276 48 46 135 128

0.044