Study setting
The present study employed a maximum difference scaling experiment to investigate consumer preferences for multinational food corporations’ products over local foods in the Greater Accra Region of Ghana, which is the most populous and urbanized among the 16 administrative regions of Ghana. The region encompasses an area of approximately 3,245 square kilometres and has an estimated population of 5,446,237, with an urban population increase of about 37.7% between the years 2010 and 2021 [34]. Given that the Greater Accra Region serves as the political capital of Ghana, it is a major economic hub that heavily influences consumer behaviour in terms of multinational food corporations’ products[35].
Study design and data collection
To conduct this study, we surveyed 200 consumers within the Greater Accra Region (Accra) over a period of three weeks in March/April 2023 using paper–pencil questionnaires. We utilized a random sampling technique to obtain the data collected through interviewer-administered questionnaire. This method aid the respondents in understanding the questions and writing out their responses. However, self-administration of the questionnaire was allowed upon request by some respondents to minimize potential interviewer bias. Respondents were approached at multinational supermarkets and international fast-food restaurants such as Shoprite Holdings Ltd, Barcelos Ghana, PICK ‘N PAY, Burger King, Massmart, Chicken Inn, SPAR, Kentucky Fried Chicken, Melcom Group, Pizza Hut, and Pizza Inn. Prior to data collection, permission to collect data was obtained in the form of written consent from the respondents after duly explaining to them the purpose of the study. Respondents were informed that their participation was voluntary, and they were at liberty to decide whether to participate or not in the study.
The survey questionnaire consisted of two sections. The first section, Section A, gathered socio-demographic characteristics of the respondents as well as reasons that inform consumers’ decision to select or choose products and/or meals from international food corporations and fast-food restaurant chains. The second section, Section B, focused on factors that influence consumers’ choice of multinational food corporations’ products over local foods. Prior to the main survey, a pilot study was conducted to identify possible challenges and problems during data collection, assess respondents’ understanding of the various factors, manage the length of the questionnaire and respondents’ reasons or basis for indicating their preferences. Adjustments were made to the questionnaire to address the respondents’ issues and enhance comprehension while reducing information overload and cognitive burden [36].
The sample for this study consisted of 200 respondents who completed the survey. It is worth noting that the sample size exceeded the minimum sample size suggested by [36] for the number of choice scenarios presented in the questionnaire. According to their proposed sample size calculation, approximately 77 respondents were needed to accurately estimate preference weights. However, the obtained sample size was about two times larger than the minimum required. Therefore, all 200 respondents were included in the final analysis as there were no missing responses.
Experimental design
Maximum difference scaling is a state-of-the-art approach for conducting consumer experiments [37]. Interest in using this method is growing in diverse areas [38] such as health [39,40,41], and environmental sustainability [42]. Researchers have discussed the potential for wider application of such experiments in food-related consumer research [43,44,45]. In this experimental design, each respondent is asked to select the most-preferred attribute and the least-preferred attribute from at least three profiles in a given choice set [46]. One of the main benefits of maximum difference scaling is its capacity to estimate the relative importance of all attributes on a common scale. Unlike traditional rating scale surveys, maximum difference scaling involves greater involvement and cognitive effort, which may help consumers focus when completing the choice task [47]. The appeal of maximum difference scaling relative to discrete choice experiments [48] has also been highlighted [43, 49].
To identify potential factors that could inform consumers’ routine decision to choose multinational food corporations’ products over local foods, an extensive literature review was conducted [15, 50,51,52,53,54], along with expert consultation involving food actors. Subsequently, a focus group discussion was conducted with 10 food actors and 30 potential consumers of multinational food corporations’ and international fast-food restaurant products. The initial list of potential factors was narrowed down to 16 plausible attributes, which are presented in Table 1.
To ensure manageable and comprehensible choice sets for the respondents, 20 choice sets were created using a balanced incomplete block design [55]. The balanced incomplete block design employed for \(k\) attributes is denoted as \(\left( {b,r,v,\lambda } \right)\) where \(b\) is the number of choice sets (blocks), \(r\) is the repetition per level, \(v\) is the number of items in each choice set (block size) and \(\lambda\) is the pair frequency. For example, the design noted as 20, 5, 4, 1 for 16 attributes has 20 choice sets, each attribute appears 5 times across all choice sets, each choice set contains four attributes, and each attribute appears once with each other. The 20 choice sets generated from the balanced incomplete block design contain four attributes per set. This approach mitigated the issue of cognitive overload and minimized the potential cognitive burden that may be induced by presenting too many attributes within each choice set [36, 56]. During the survey, each participant was presented with the 20 choice sets, with each set comprising four attributes, as depicted in Fig. 1. The respondents were required to express their preferences by selecting the “best” (most important) reason (attribute) and the “worst” (least important) attribute while considering purchasing a multinational food corporations’ product over local food (when there is a means or an option to eat local food) related to the situation described in Fig. 1. The situation was defined to standardize the reasons for considering purchasing a multinational food corporation’s product over local food and to avoid confusion with special situations where people might think about directly comparing preferences for multinational food corporations’ products to local foods as frequently encountered in discrete choice experiments [48], where respondents have to compare product descriptions and select one alternative in a choice set.
A sample completed maximum difference scaling experiment choice set as presented to respondents
Empirical strategy/ Data analysis
In a maximum difference scaling experiment, profiles are evaluated using a random utility framework [57, 58]. The choice frequencies for best and worst options in a choice set are used to compare the relative importance of different attributes. The Maximum difference model estimates the underlying utility of each choice.
To formalize this model, we denote \(\hbar\) with \(\left| \hbar \right| \ge 3\) as the finite set of potentially available options from a choice set and let \(\psi \left( \hbar \right)\) denote the statistical experimental design, that is, the set of (sub)sets of choice options that occur in this study. For any set \(Y \in \psi \left( \hbar \right),\)
\(Y \subseteq \hbar\) with \(\left| Y \right| \ge 3\), let \(P_{Y} \left( i \right)\) and \(P_{Y} \left( j \right)\) denote the probability that respondents select a pair of items \(i\) and \(j\) from set \(Y\), where \(i\) is selected as the best and \(j\) is selected as the worst, and the difference in utility between the two items is the maximum among all utility differences. Here \(P_{Y} \left( {i,j} \right)\) is the probability that the item \(i\) is selected as the best and item \(j \ne i\) is selected as the worst.
By assuming that there is a scale \({\upmu }\) such that for all \(i \in Y \in \psi \left( \hbar \right),\) where the value \(\upmu \left( i \right)\) for an item \(i\) is interpreted as the utility for that option, the best choice model can be formulated as
$$P_Y\left(i\right)=\frac{e^{\mathrm\mu\left(\mathrm i\right)}}{\sum_{Z\in Y}e^{\mathit\mu\mathit{\left(z\right)}}}$$
(1.1)
The parallel worst choice model can be reformulated as follows if we assume that there is a scale v such that for all \(j \in Y \in \psi \left( \hbar \right),\)
$$P_Y\left(j\right)=\frac{{e}^{v\left(j\right)}}{\sum_{Z\in Y}e^{v\left(z\right)}}$$
(1.2)
If both the corresponding choice probabilities on best and worst item satisfy all distinct pairs \(i,j \in Y \in \psi \left( \hbar \right),\) then
$$P_{{\left\{ {i,j} \right\}}} \left( i \right){ = }P_{{\left\{ {i,j} \right\}}} \left( j \right),$$
and we obtain
$$P_Y\left(j\right)=\frac{e^{-\mathrm\mu\left(\mathrm i\right)}}{\sum_{Z\in Y}e^{-\mu\left(z\right)}}$$
(1.3)
Assume that the choice probabilities satisfy the corresponding best and worst model, and that the utility of a choice alternative in the selection of a best option is the negative of the utility of that option in the selection of a worst option, and this utility-scale \(\upmu\) is such that for all \(i,j \in Y \in \psi \left( \hbar \right),\)
\(i \ne j,\)
$$P_{Y} \left( {i,j} \right) = \frac{{\mathop e\nolimits^{{\left[ {\upmu \left( i \right){ – \mu }\left( j \right)} \right]}} }}{{\sum\nolimits_{{\left\{ {p,q} \right\} \in Y}} {\mathop e\nolimits^{{\left[ {\upmu \left( p \right){ – \mu }\left( q \right)} \right]}} } }},$$
(1.4)
where \(\upmu \left( i \right)\) is the systematic component of the utility of item \(i,\) which is assumed to be \(\upmu \left( i \right) = \beta_{i} X_{i}\), where \(\beta_{i}\) is a preference coefficient to be estimated and \(X_{i}\) is a dummy variable taking the value 1 if item \(i\) is included in a choice set, and 0 otherwise. In this study, consumers independently select the attributes related to multinational food corporations’ products/meals they like and dislike the most when compared to local meals.
We fitted the maximum difference model to our data using JMP Pro Version 16.0. Statistical significance was measured at p-values of less than 0.001, 0.01, and 0.05. In the absence of p-values, statistical significance was measured at 95% confidence intervals (CIs) greater than or less than zero. A significant positive/negative preference coefficient indicates a high/low preference for a specific attribute. The sign of the preference coefficient indicates whether the plausible attribute has a positive or negative effect on utility. We compared the relative importance of the different attributes across attributes given the utility estimates (preference coefficients).
Add Comment