Evaluating the framing bias in the supplier selection process

Evaluating the framing bias in the

supplier selection process

A vignette-based experiment on the framing bias and the influence of uncertainty in the supplier selection process.

Master’s Thesis Double Degree in Operations Management Frank Hazeborg S2376997 and B6068194 11 December 2017 Supervisors: dr. W.M.C. van Wezel dr. Y. Yang

MSc Technology and Operations Management Faculty of Economics and Business

University of Groningen

MSc Operations and Supply Chain Management Newcastle University Business School

1

Abstract

3

Acknowledgements

Special thanks goes to my supervisors, without whom this project would not be possible. First, I would like to thank dr. W.M.C. van Wezel for supervising this thesis. I am particularly grateful for your help and insights during our feedback sessions, that helped me tremendously. I would also like to thank dr. Y. Yang to supervise me from Newcastle, our Skype sessions were of great value to me.

I would also like to thank everyone that participated in this research. Without you, this research would not have been possible. Thank you for your time and effort.

4

Table of content

The role of uncertainty in supplier selection decisions ... 11

Methodology ... 15

Choice of research design ... 15

Design ... 16 Pre-design stage ... 16 Design stage ... 17 Post-design stage ... 18 Sampling strategy ... 19 Vignette-based experiment ... 21 Attention check... 22 Manipulation check ... 23 Additional Statements ... 23 Procedure ... 26 Ethics ... 26 Results ... 27 Descriptive statistics ... 27 Manipulation check ... 29 Hypothesis 1 ... 29 Hypothesis 2 ... 30 Additional Statements ... 32 Post Hoc ... 34 Discussion ... 36 Theoretical implications ... 36

The framing effect ... 36

5

Appendix I - Hypothesis 1 ... 51

Appendix II – Hypothesis 2 ... 54

Appendix III – Attention check ... 55

Appendix IV – Additional statement analysis ... 56

Appendix V – Post hoc analysis ... 61

Appendix VI - Effect of uncertainty within framed groups. ... 66

Appendix VII – Vignette-based experiment ... 68

6

Introduction

Traditionally, supply chain management activities, such as supplier selection decisions, are based on models which are a simplified view of reality (Boudreau et al., 2003). These models are mainly based on the assumption of ‘homo economicus’, with rational and analytical decision-making at its core (Carter et al., 2007). While there is a wide range of knowledge regarding such decisions, people often tend to overlook the influence of human behaviour (Tokar, 2010). Due to behavioural issues, people systematically do not make choices in accordance with normative or optimal policies. Especially in judgement and decision-making, behavioural issues are frequently occurring (Bazerman, 2002). Therefore, behavioural aspects need to be included in supplier selection models and activities. Otherwise the robustness, predictive accuracy and overall value of supplier selection models and activities will be limited (Tokar, 2010).

In supply chain management literature, a broader recognition of understanding the role of human behaviour in supply chain management has begun (Bendoly et al., 2006, 2010; Gino and Pisano, 2008). As our understanding enlarges, it becomes clear that the role of human behaviour in supply chain decision-making has positive and negative aspects. Blattberg and Hoch (1990) stated that human judgement and decision-making can add extra value when decisions cannot be completely captured by models. However, human decision-making can also be significantly distorted by various biases (Carter et al., 2007). Therefore it’s important to understand which, to what extent and under which circumstances biases can affect supply chain decisions. Especially under conditions of uncertainty, biases can systematically affect decision-making (Kahneman, 2012; Tversky and Kahneman, 1974). As supplier selection decisions often cope with a high degree of uncertainty (Kull et al., 2014), research on biases in this particular area can be fruitful. More specifically, Carter et al. (2007) suggest an evaluation of uncertainty as a moderating effect on decision biases in supplier selection decisions, as a direction for further research.

7

Our research will focus on the role of the framing effect under different degrees of uncertainty in the supplier selection process. The moderating effect of uncertainty on the framing effect has not been studied yet. Therewith, this research will contribute to the field of behavioural supply chain management. Furthermore, this research is practically relevant to managers or decision-makers involved with supplier selection decisions. On both sides of the transaction, suppliers and buyers can benefit from this research. Both suppliers and buyers need to be aware of the framing effect and in which situations the framing effect will have the most impact on decision-making. Buyers benefit by being able to make decisions more objectively. Suppliers benefit by being able to choose the best wording to present their information in the most effective manner.

In this research, the moderating effect of uncertainty on the framing effect is investigated. By conducting a two-by-two between subjects vignette-based experiment, this effect is tested. Professionals with supplier selection experience are asked to rate a supplier in a hypothetical scenario where certain supplier characteristics are framed positively or negatively. For each type of framing, the degree of environmental uncertainty can be either high or low. With this research, we try to answer the following research question:

To what extent does the framing effect influence supplier selection decisions under different degrees of uncertainty?

This research is structured as follows. First the theoretical background of supplier selection, the framing bias, uncertainty in supply chains and our posed hypothesis to answer our research question will be discussed. Second, we will discuss how the vignette-based experiment was designed and conducted. Third, we will present our results. The fourth section will provide an interpretation of the results, together with theoretical and managerial implications and the limitations of our research. Finally, in the conclusion section we revisit our posed hypotheses and discuss directions for further research.

Theoretical background

8

In the supplier selection process, the first step is ‘supplier scoring and assessment’(Chopra and Meindl, 2013, p. 441). In this step, suppliers will be rated on their performance on multiple areas. In practice, managers often base their decisions solely on the price of the goods of a certain supplier. However, many other supplier characteristics, such as lead time, quality, on-time delivery and many other characteristics also affect the total cost of doing business with a certain supplier (Chopra and Meindl, 2013, p. 441).

A well performed supplier scoring and assessment results in a supplier that is scored along all dimensions and evaluates the impact on total cost of doing business with the supplier (Chopra and Meindl, 2013, p. 441). In the supplier selection process, the output of the supplier scoring and assessment is used to select the appropriate supplier (Chopra and Meindl, 2013, p. 441). The supplier selection process is a critical step that directly influences production costs, product quality and lead times (De Boer et al., 2001; Cousins and Spekman, 2003; Kannan and Tan, 2002; Murray et al., 2005), therefore supplier selection is of great importance in to the firm. The rational and analytical supplier selection process is treated as an economic decision by supply chain management literature, focusing on minimal cost incurred in buying the product (Mantel et al., 2006). This approach is anchored in ‘transaction cost economics’ (Williamson, 1981). Literature assumes a rational decision approach made by the firm, where an optimal rule-based and programmed decision is made given the known factors and uncertainty dimensions (Mantel et al., 2006). According to Mantel et al. (2006) there are two shortcomings to this approach. First, it is typically not the firm that makes a decision. Instead, a human being makes the decisions in the firm. As a result, there is a strong human component within corporate decision making. Second, the human decision maker is likely to involve other information and cognitive biases in the rule-based decision approach.

9 Framing Bias

Carter et al. (2007) already introduced a set of judgement and decision-making biases to the field of supply chain management. The overview presented by Carter et al. (2007) lists the biases that could possibly have an effect on decision-making in supply chain management. The Framing effect has been proposed as a cognitive bias that could significantly affect decisions in supply chain management (Schultz et al., 2017).

Kahneman and Tversky's (1979) prospect theory is the most influential explanation of the framing effect. Tversky and Kahneman (1981) clarify this concept with their groundbreaking ‘Asian disease problem’. Participants in this experiment could solve this problem by choosing one out of two options. The participants could choose from a deterministic option or a gamble with the same expected value. When the problem was worded in terms of gains, participants were risk averse and a great majority chose the deterministic option. When the problem was worded in terms of losses, participants were risk-seeking and the majority chose for the gamble. This phenomenon is known as risk reversal due to framing (Schultz et al., 2017). This specific type of framing was later classified as risky choice framing (Levin et al., 1998).

Based on an extensive literature review, Levin et al. (1998) argue that different operational definitions of framing have tapped different underlying processes and that “All frames are not created equal”, which means that there is more than one type of framing. Besides risky choice framing, Levin et al. (1998) also classified goal framing and attribute framing. Follow up studies have shown that all three types have reliable effects and are independent from one another (Levin et al., 2002).

10

The last, and most simple type of framing is attribute framing (Levin et al., 1998). Attribute framing is an elementary form of framing because single attributes can be manipulated, by wording the same attributes differently. E.g. Levin & Gaeth's (1988) classical example of ground beef: a hamburger can be advertised as 25% fat or 75% lean). Compared to risky choice and goal framing, attribute framing is very simple, it allows for the most straightforward testing of the influence of the framing effect (Levin et al., 1998). Attribute framing can be applied in many types of information prior to a decision, as long as there is at least one attribute suitable for framing. In decision-making involving attribute framing the focus is not on the choice between a certain and a risky option, as is risky choice framing. Instead, the outcome of the decision is typically a measure of favourability, such as a numerical rating (Tokar et al., 2016). The effect of attribute framing is strong and consistent across a wide variety of decision contexts and performance evaluations (Tokar et al., 2016). As the supplier selection process is mainly based on evaluating a suppliers’ performance (Wouters et al., 2009), it is likely that decision-making in this area will be susceptible to the framing bias. Supplier selection criteria (or supplier attributes) are often suitable for framing. For example, the delivery rate of a certain supplier can be positively framed as “delivering 19 out 20 times on-time” vs. negatively framed “delivering 1 out of 20 times too late”. Both framings present the same attribute in a different wording.

To stress the differences between the three different types of framing, the following table adopted from Levin et al. (1998) is included below.

Table 1 – Summary of different types of framing (Levin et al., 1998)

Frame type What is framed What is affected How effect is measured Risky Choice Set of option with

different risk levels

Risk Preference Comparison of choices for risky options

Goal Consequence of a

behaviour

Impact of persuasion Comparison of rate of adoption of behaviour

Attribute Object/event

attributes or characteristics

11

To counteract the framing effect, there are many objective supplier selection matrixes, decision models and measures available to make rational objective decisions (Chopra and Meindl, 2013). However, it is evident in the field of cognitive science that the framing effect can have a high impact on human decision-making. Kahneman & Tversky (1981) already showed that professionals are susceptible to the framing effect. The effect can have a high impact on a professional’s decision-making in all kinds of industries and is still widely accepted in literature (Freling et al., 2014; Kull et al., 2014; Tokar et al., 2016).

Although, there are also situations where the framing effect is less likely to affect the decision. For example, in situations that involve strongly-held attitudes and beliefs (e.g. abortion decisions) or high personal involvement (Levin et al., 1998). No scientific evidence is found that supplier selection decisions involve strongly-held attitudes or beliefs or high personal involvement.

Tversky and Kahneman (1974) already stated that decision-making can be systematically affected by biases, especially under conditions of uncertainty. Up until today, their theory is relevant and used by many authors (Kaufmann et al., 2009, 2012; Tokar, 2010). As supplier selection decisions are decisions that often cope with uncertainty (Kull et al., 2014), the framing of supplier attributes may have a strong effect on supplier assessments. Therefore, the first hypothesis is as follows:

Hypothesis 1: Supplier selection professionals are susceptible to the framing bias.

The role of uncertainty in supplier selection decisions

12

There are multiple contextual factors that influence the uncertainty in the decision-making context (Kull et al., 2014). In organization theory, Milliken's (1987) measures are often used to describe uncertainty in decision-making (Birasnav et al., 2015; Chen et al., 2016). Following Milliken (1987), the uncertainty enlarges when the decision-maker is less able to predict future events in the environment. Besides, the inability of predicting accurately what the consequences of the decision might be is typical for decisions with high uncertainty. Thereby, Birasnav et al. (2015) state that in supply chain situations, high uncertainty can be characterized by unstable and unpredictable technologies, suppliers and customers (Birasnav et al., 2015).

Lee (2002) characterizes different products and processes in four categories with different uncertainties in supply chains. Demand uncertainty is linked to the predictability of the demand for a specific product. Functional products are products that have a stable demand and long product life cycles, while innovative products have unstable demand, short product life cycles with high innovation and often fashion trends. On the supplier side, other kinds of uncertainties occur. A stable supply process, with low uncertainty involves mature technologies and the supply base is well established. An evolving supply process, with high uncertainty involves manufacturing processes and technologies that are not mature yet. The processes and technologies are under development and are rapidly changing. Suppliers may be of limited size and may be inexperienced, as they go through innovations themselves. Lee’s (2002) framework can be seen in figure 1.

Figure 1- Supply chain uncertainty matrix, Lee (2002)

13

supplier selection decisions in this category might be highly susceptible to biases. As their theory states that biases particularly occur in situations of uncertainty. On the opposite, supplier selection decisions in a basic apparel supply chain are made in a low uncertainty context. Here, the supply base is experienced, technologies in the supply chain are matured and the demand can be forecasted with high accuracy. Supplier selection decisions in this category might be less susceptible to biases according to this theory. In this way, the uncertainty of the decision context in supplier selection decisions could moderate the effect of the bias. Following this theory, the framing effect might be stronger for supplier selection decisions for a semiconductor firm compared to supplier selection decisions for a basic apparel firm.

Furthermore, literature suggests other explanations for a stronger framing effect under uncertainty. First, decision-makers could be limited by cognitive constraints to process available information under uncertainty (Kahneman, 2003; Kaufmann et al., 2009; Simon, 1990). Wouters et al. (2009) describe how the need for processing information increases when uncertainty increases. However, research underlines that decision-makers are limited by cognitive constraints (Kahneman, 2003; Simon, 1990). As a result, decision-makers are likely to reduce their information processing efforts if the amount of processing is beyond their cognitive ability (Koufteros et al., 2002). With reduced information processing efforts, decision-makers are more susceptible to cognitive biases, and might result into ‘satisficing’ (Simon, 1957, p. 204) rather than rational decision-making (Kaufmann et al., 2009).

Second, Pitz and Sachs (1984) have stated that biased decision-making can function as a guide to the final decision in environmental uncertain situations. In uncertain situations, where better comprehensive and cautious decision-making is not possible, biases and heuristics may provide an effective way to make a decision by approximation (Haley and Stumph, 1989; Kahneman, 2012; Tversky and Kahneman, 1974). Under uncertainty, people do not rely on statistical methods. Instead, they rely on heuristics which can lead to systematic errors (Kahneman and Tversky, 1973). People often rely on intuition to suppress uncertainty (Lipshitz and Strauss, 1997). This intuitive processing can lead to suboptimal decisions and biases (Kahneman and Klein, 2009; Lee et al., 2009). Therefore, the more uncertain a decision context is, the higher the framing effect might be.

14

Lipshitz and Strauss, 1997). Research states that gathering additional information is an important strategy to reduce uncertainty in the decision context (Lipshitz and Strauss, 1997; Wouters et al., 2009). Therefore it can be expected that purchasing managers will try to make an optimal decision by taking all information into account and make a well-informed decision when confronted with uncertainty. This deliberate way of decision-making is expected to reduce the framing effect. In this way, the framed attributes have less impact on the final decision.

The second tactic is to reduce uncertainty by soliciting advice or opinions from experts, colleagues, etc. in making (Lipshitz and Strauss, 1997). With this tactic, the decision-maker will base its decision on advice or opinions from others. This tactic can be compared with the ‘adjustment from the starting point’ heuristic described by Tversky and Kahneman (1974). According to Tversky and Kahneman (1974), people use this heuristic in numerical predictions. Here, one’s judgement is based on a starting point, like an initial value that is given. The adjustment from the starting point is typically insufficient, as the final estimate remains close to the starting point (Tversky and Kahneman, 1974). Due to insufficient adjustment from the starting point, the heuristic fails to produce the right judgement. In this way, using the heuristic leads to an anchoring bias in decision-making (Kahneman, 2012).

The anchoring bias describes how an initial piece of information can hold significant sway over decision-making, to such extent that this piece of information is overqualified and other available information is underqualified (Botros et al., 2014). Therefore, the soliciting advice tactic may reduce the framing effect, given that advice from colleagues is present and environmental uncertainty is relatively high (as biases typically occur under uncertainty (Bazerman, 2002; Kahneman, 2012)). Important here is to note that the anchoring bias is not an uncertainty reduction tactic. The anchoring bias affects decision-making as a result of a heuristic that fails to produce the right judgement.

Third, uncertainty can be reduced by relying on formal and informal rules of conduct, such as norms and ‘standard operation procedures’. Supplier selection professionals can quantify uncertainty and use formal schemes to reduce uncertainty (Lipshitz and Strauss, 1997). This objective way of reasoning may reduce uncertainty and thereby the framing effect.

15

2013). This form of uncertainty is expected to affect biased decision-making. The premise is that a higher degree of uncertainty in the decision context will increase the framing effect. Therefore, the second hypothesis is:

Hypothesis 2: The degree of uncertainty in supplier selection moderates the framing effect, such that the framing effect will be stronger as uncertainty increases.

Figure 2 – Conceptual Model

Methodology

In this research, a vignette-based experiment was designed and conducted by combining aspects of traditional experiments and surveys. There are four basic elements that are critical to a traditional experimental design, regardless of research context: (1) the participants are randomly selected, (2) the participants are randomly assigned to different treatment conditions, (3) the experimenter manipulates treatments in the experiment and (4) the experimenter has control over the conduct of the experiment (Bachrach and Bendoly, 2011).

16

On the other hand, a traditional survey does show high external validity. This is due to the high representativeness, multivariate and multivalent measurements. However, traditional surveys have low internal validity, because of multicollinearity of measured variables and the experiment is executed in a passive way (i.e. the experiment is executed without any experimental intervention or control of explanatory variables) (Atzmüller and Steiner, 2010). By combining the traditional survey with a traditional experiments, by conducting a vignette-based experiment, the external and internal validity can be improved (Atzmüller and Steiner, 2010). Vignette-based experiments have high internal validity, due to a high stimulus uniformity and high degree of control over the stimulus (Hora and Klassen, 2013). As the researcher manipulates each treatment by manipulating each type of vignette, the participants get different stimuli. This active type of research is able to measure causality, where surveys measure correlations. The active way of measuring and high stimuli control reduces multicollinearity1 in vignette-based experiments (Atzmüller and Steiner, 2010; Hora and Klassen, 2013). Furthermore, the external validity will be higher for vignette-based experiments due to the representativeness of the sample. Vignette-based experiments allows researchers to more efficiently access professionals (Knemeyer and Naylor, 2011), for example through online channels.

Design

To test the hypothesis that are posed, a based experiment was conducted. A vignette-based experiment is practically a survey, with a vignette as its basic item. The vignette is a card with a short description about a hypothetical scenario. This scenario is generated by the researcher by manipulating characteristics across different versions of the vignettes (Ganong and Coleman, 2006). In our research we create a hypothetical supplier selection scenario in which the purchasing manager has to rate a given supplier (‘supplier X’) to put it on an approved vendor list or not. The organizational environment will either have high or low uncertainty and the supplier characteristics will be either positively framed or negatively framed. Our vignettes are adopted from the vignette-based experiment conducted by Kull et al. (2014).

Pre-design stage

The vignette-based experiment is designed according to the three stages of vignette creation and validation (Rungtusanatham et al., 2011). According to Bendoly (2015, p. 28) the ‘three

1 _{Note that this does not apply to additional survey questions (to measure underlying mechanisms) included in}

17

stages’ of Rungtusanatham et al. (2011) is a “very good source” for the experimental design of a vignette-based experiment. Furthermore, the three stages are suitable to investigate behavioural aspects in supply chain management (Knemeyer and Naylor, 2011).

The first stage is to “get to know the context” and the intent of this stage is twofold (Rungtusanatham et al., 2011). First, the researcher should get to know the context of the decision-making that is investigated. Second, the researcher should “get to know the factors of interest” at measurement level. The researcher should determine in this stage which factors need to be manipulated between the different versions of the vignettes (Rungtusanatham et al., 2011). In the pre-design stage, first the context of a supplier selection decision was analysed. This stage is used to get familiar with supplier selection decisions and uncertainty in supply chains. This information will later be used to create a realistic hypothetical decision-scenario. Second, the factors of interest were analysed, these factors were derived from literature research. In our research, supplier characteristics were either framed positively or negatively and the decision context was manipulated to either have high or low uncertainty between vignettes.

Design stage

The design stage is the “structured creative writing stage” (Rungtusanatham et al., 2011). The intent of this stage is to write multiple version of the vignettes in which the factors of interest vary between the vignettes. Each vignette should consist of ‘common modules’ and ‘experimental cues modules’ (Rungtusanatham et al., 2011). Each common module consists of contextual information about the scenario that is invariant across the different versions of the vignette. Each experimental cues module consists of information that is variant across the different versions of the vignette (Rungtusanatham et al., 2011).

18 Table 2 – Classification of vignettes

Positive Framing Negative Framing

Low Uncertainty Vignette 1

Industry: White T-shirts

“Fraction of on-time delivery: 19 out of 20”

“Sufficient quality: 94%”

Vignette 2

Industry: White T-shirts

“Fraction of too late delivery: 1 out of 20”

“Insufficient quality: 6%” High Uncertainty Vignette 3

Industry: Computer chips

“Fraction of on-time delivery: 19 out of 20”

“Sufficient quality: 94%”

Vignette 4

Industry: Computer chips “Fraction of too late delivery: 1 out of 20”

“Insufficient quality: 6%”

Each vignette starts with an experimental cues module. In this module, the participants get to know their role and the task they have to perform in the hypothetical scenario. The focal firm of the vignette is variant between the high and low uncertainty vignettes, to match a real-world scenario. This module is followed by an experimental cue that describes the environmental context in terms of uncertainty in the supplier selection decision. The common module that comes after describes how the supplier evaluation task should be performed.

Another common module describes characteristics of supplier X, the hypothetical supplier that needs to be evaluated. In this module, the time that supplier X is on the market is variant between high and low uncertainty vignettes to make the vignette more realistic.

The vignette ends with an experimental cue module where the framing of the participant happens. This module contains extra information about supplier X. These text fragments are either positively or negatively framed between different vignettes. The framed text fragments are listed in table 2.

Finally, each vignette contains a response item to measure the effect of the manipulations (Bendoly and Eckerd, 2013). In our vignettes, the response item is the supplier evaluation score. Each vignette can be found in appendix VIII.

Post-design stage

19

vignette needs effective experimental cues, such that the participant receives the desired levels of manipulated factors of interest (Rungtusanatham et al., 2011).

In the post-design stage, a pilot test was used to check for mistakes and to improve the vignettes. First, the low uncertainty vignette was initially based on a hypothetical milk processing firm. When participants were asked about their way of reasoning, they seemed to have strongly held attitudes about the characteristics of ‘milk supplier X’, regardless of positive or negative framing. Strongly held attitudes about the decision can reduce the framing effect (Levin et al., 1998). Using this type of industry could reduce the effectiveness of the vignettes. Therefore, we decided to change the focal firm to a firm that produces basic apparel (e.g. white T-shirts). Second, we checked for clarity of the vignette. With this check, we found out that the uncertainty between the high and low uncertainty vignettes was not perceived well. Without perceived differences in uncertainty, we were not able to test our second hypothesis. Therefore, we have decided to repeat the information about the uncertainty (“In the… very (un)stable”) in the last experimental cue module. In this way, the information about the uncertainty in the decision context is mentioned twice in the vignette.

The third problem we encountered was a high amount of variance in supplier evaluation scores within each experimental group. This was most likely caused by the lack of a ‘starting point’ or ‘anchor’ to rate supplier X. When an initial starting value is given in decision-making, people are likely to adjust their rating from this number (Bendoly et al., 2010). In this way the scores are likely to get more concentrated around the starting point. If the variance would not have been reduced, our sample size should have been considerably larger before we could draw conclusions from our results. To solve this problem we have added an extra cue with ‘starting points’ to decrease variance. We have done this by adding “Other colleagues have rated similar suppliers Y and Z with respectively 70 and 74%.” to each vignette. This extra common module reduced the high amount of variance within each experimental group to an acceptable level and is invariant across the different versions of the vignette.

Sampling strategy

20

initially been used for pilot testing (together with pilot testing on students) in the post-design stage.

21 Vignette-based experiment

Vignette 1 can be found below. Vignette version 1 is characterized by a low uncertainty context and positively framed supplier attributes (vignette 2, 3 and 4 can be found in appendix VIII). One of the four vignettes is randomly assigned to a participants. The vignettes were designed by the three stages of Rungtusanatham et al. (2011) and adopted from (Kull et al., 2014). The response item was included at the bottom of each vignette. Here, participants could fill in their supplier evaluation score.

V1. PLEASE READ THE CASE CAREFULLY,

You are a purchasing manager for a medium sized firm that produces basic clothing, like white T-shirts. The firm is one of the oldest clothing manufacturers in the Netherlands. As the top-management seeks to increase the firm’s performance, you are asked to evaluate several suppliers on their performance. The suppliers that need to be evaluated are suppliers of the main element of clothing, namely textile. The supply base is well established and the technologies that are used in the supply chain have been fully matured. Furthermore, we know that the demand for basic clothing is very stable and can be forecasted with high accuracy. The top-management asks you to evaluate supplier X and advise whether this supplier should be put on an approved vendor list (i.e. a list from which a supplier can be selected later). A score of 0% indicates that you think that the supplier should definitely not be on the approved vendor list, whereas a score of 100% indicates that the supplier definitely should

be put on the approved vendor list. . Supplier X is an internationally operating textile firm that operates in the textile market for

multiple decades. Supplier X claims to be a company that produces textile only with the highest quality of cotton. According to supplier X, they are able to deliver textile slightly below

the market price due to economies of scale. . You decide to ask your purchasing team, which helps you to keep track of supplier

performance indicators. A colleague says the following:

‘’We already know that this supplier delivers 19 out of 20 of their deliveries on time. When the textile arrives at their customers, the textile is of sufficient quality 94% of the time. In the clothing industry, suppliers of textile are normally very developed and demand for our clothing is very stable.’’.

Other colleagues have rated similar suppliers Y and Z with respectively 70 and 74%. Although the information provided in the scenario above is limited, please answer the following question:

22 Attention check

This research heavily depends on the goodwill and comprehension of the participants. As the vignette-based experiment is conducted through online channels, our ways to examine if participants used thoughtful or non-thoughtful processes in our vignette-based experiment are limited. E.g. participants could be occupied with other tasks, loose focus, lack comprehension, etc. during the experiment.

The online channels used in our vignette-based experiment can be roughly divided into two categories. The first category of supplier selection professional has been approached via personal networks and Purchasing networks on LinkedIn. The second category has been approached via Amazon Mturk. The Amazon Mturk platform enables researchers to obtain high-quality data fast and at low cost (Buhrmester et al., 2011). The downside of this platform is that there are also participants with the wrong intentions. To filter out unwanted participants, attention check questions were used in combination with a filter based on purchasing experience and the time it took the participant to complete the experiment.

First, participants were filtered out based on their purchasing experience. Participants that had no purchasing experience were filtered out for further analysis. The second filter was based on the duration to complete the vignette-based experiment. Taking into consideration that experienced Amazon Mturk users can process information rather fast, the rejection threshold based on time was set to two minutes or less. Third, if it took a participant more than two minutes to complete the vignette-based experiment, the case was individually approved or rejected based on three attention check questions. Recent studies have shown that that various forms of attention check questions can increase the quality of the data collected on Mturk (Peer et al., 2014). The three attention check questions were used to check whether a participant did read the vignette thoroughly. The rejection or approval of each participant depends on their answers (5 point Likert scale from strongly disagree to strongly agree) to the following three statements:

I think that the environment in the business context of the firm was uncertain. (Q1) The demand for <product in vignette> was very stable. (Q2)

The supply base of <raw material for product in vignette> was developed. (Q3)

23

case of vignettes 1 and 2, the participant’s answers were accepted as valid when the average of the scores on statement 1 (reversed), 2 and 3 indicated a higher score than neutral. In case of vignettes 3 and 4, the participant’s answers were accepted as valid when the average of the scores on statement 1 (reversed), 2 and 3 indicated a lower average score than neutral. The first question is relatively subjective, the second and third question could be literally derived from the vignette. Therefore the deviation from neutral for question 1 was multiplied by 0.5 in the calculation.

Manipulation check

Two extra statements were included in the vignette-based experiment, to investigate if participants perceived uncertainty differently between differently manipulated vignettes. In contrast to the attention check questions, these statements could not be literally derived from the text. An independent samples t-test was conducted to see if the scores on statement Q4 and Q5 significantly differ between the high and low uncertainty groups. The statements were derived from Milliken (1987) and Riedl et al. (2013) and are as follows:

I had too little information to rate supplier. (Q4)

I felt able to predict accurately what the consequences of my rating might be, for the <focal firm of vignette>.(Q5)

Additional Statements

Based on literature, expectations about the underlying mechanisms of our findings were formed. In our theoretical background section, possible underlying mechanisms were already mentioned that could explain our findings. These possible underlying mechanisms are based on our expectations derived from literature. These underlying mechanisms are not included in our conceptual model and are therefore not a primary part of our research. Items that measure the underlying mechanisms are included for exploratory purposes and can be used as directions for further research. The items do not measure what the effects are of the framing effect and uncertainty (as a moderator) on the supplier evaluation score. Instead, the items try to predict why uncertainty moderates the framing effect. In this section, we will discuss how we have measured the underlying mechanisms and how this relates to literature.

24

vignette on their importance (on a five-point Likert scale) to rate supplier X. This is done with the following question: “To what extent did you find the following information important to rate supplier X?” followed by seven text fragments. The text fragments contain information of supplier X that could have been derived from the vignette. Again, this question is also added for exploratory purpose.

First, we try to investigate if participants used uncertainty reduction tactics. As explained in our theoretical background, these tactics may affect the strength of the framing effect under different degrees of uncertainty. Kaufmann et al. (2009) and Lipshitz and Strauss (1997) suggest that participants gather additional information to reduce uncertainty. This tactic was measures with statement Q6 (derived from Lipshitz and Strauss (1997)).

I tried to take as much information into account as possible to make a good decision. (Q6)

This tactic was further investigated by investigating the ratings of importance of all seven text fragments. In this way, we are able to find out more about which information is gathered under different degrees of uncertainty. The seven text fragments are listed below:

Supplier X in internationally operating. (TF1) Supplier X uses high quality products. (TF2)

Supplier X delivers slightly below market price. (TF3) Supplier X delivers <… out of … on time/too late>. (TF4) Supplier X delivers <…% (in)sufficient> quality. (TF5) The quality ratings of other colleagues (70 and 74%). (TF6) Supplier X is on the market for … years. (TF7)

Another tactic we would expect to reduce uncertainty is soliciting advice (from colleagues etc.) (Lipshitz and Strauss, 1997; Tversky and Kahneman, 1974). As explained in our post-design section, a common module that contained advice from colleagues was added to each vignette: “Other colleagues have rated similar supplier Y and Z with respectively 70 and 74%”. We tried to investigate this tactic by asking participants about the importance of the colleague ratings to rate supplier X by using text fragment TF6.

25

Strauss, 1997). In supplier selection decisions, this would be rational decision-making based on ‘homo economicus’ (Carter et al., 2007). In our research, we try to investigate this tactic by using statement Q14 and Q15. By asking participants if they have used calculations/statistics or purchasing logic/rules/matrices/schemes to rate supplier X.

I used Calculations/statistics to rate supplier X. (Q14)

I used professional purchasing logic/rules/matrices/schemes to rate supplier X. (Q15)

Further, other variables have been included in our research to investigate underlying mechanisms other than uncertainty reduction tactics. Koufteros et al. (2002) suggest that decision-makers are likely to reduce their information processing efforts if the amount of processing is beyond their cognitive ability. In our research, reduced effort was measured by statement Q6 and Q7. Statement Q8 and Q11 measured the cognitive effort. Pitz and Sachs (1984) suggest that biased decision-making can function as a guide to the final decision in environmental uncertain situations. The way of thinking (rational vs. intuition) was measured with statement Q16 and Q17. These statements are listed below.

I tried hard to make an optimal decision. (Q7)

I found it hard to rate supplier X’s performance. (Q8)

I think the task to rate supplier X was mentally demanding. (Q11) I used rational reasoning to rate supplier X. (Q16)

I used my feelings/intuition to rate supplier X. (Q17)

Last, three statements were added to this section to control for possible factors that could obstruct the framing effect. According to Levin et al. (1998) these factors are strongly-held attitudes and high personal involvement in decision-making. Respectively, these factors were measured by statement Q12 and Q13. Furthermore, we asked participants whether they were familiar with the framing effect (Q10). If a participant would know that the framing effect is used in the vignettes, this could affect the results.

26 I am personally involved with <industry of vignette>. (Q13)

Procedure

The experiment will be executed by the use of online research survey software Qualtrics. A link to enter the survey can be shared digitally to the participants by the use of social media, email, Amazon Mturk or otherwise. Once the supplier selection professional is contacted and willing to participate, they are briefly informed about the research. They are informed about the assessment of a hypothetical scenario about a supplier selection decision.

First a general survey is done to gather the necessary supplier selection related background information about the sample group. The items consists of: years of purchasing experience (experience), average time spend on purchasing per month (Purchasing_time), size of the organisation (Nr_employees) and whether or not there is a (separate) purchasing department in the organisation (Separate_dpt). After that, the assessment of the hypothetical scenario in the vignette starts. Once the participant has assessed the vignette scenario, the participants has to agree or disagree (on a 5 point Likert scale) with multiple statements about the way of reasoning and the importance of information in the vignette.

Once one of these steps is completed, participants cannot go back to previous steps to alter information. Further, all questions are mandatory to fill in, participants cannot continue to the next step if not all questions are answered and all steps are completed.

Ethics

27

Results

Descriptive statistics

After the data collection, 611 results were ready for analysis. Of these results, 330 were deleted. These participant were either deleted because they failed the attention check, did not have professional purchasing experience or purely based on time (when they completed the study in less than two minutes).

After three additional far-outliers were removed (data point 14, 81 and 86 in figure 3) because of unnatural answering patterns, the data set consists of 278 results. From 278 participants, 173 were male and 105 were female. The average age was 34.0 with a standard deviation of 10.4. The participants were highly educated on average. 136 Participants had a Bachelor’s degree, 76 had a Master’s degree, 11 participants were professors, 2 participants had a doctorate degree and the other participants were relatively lower educated. The most occurring participant jobs are technical or engineering jobs (n=51), general managers (n=38), administrative jobs (n=28), operations/supply chain specialists (n=27) and data analysts (n=22). In the table below, the supplier selection related characteristics of the participants can be found. The names of the variable that are used in the analysis are also provided in italics.

Table 3 – Supplier selection related characteristics

Characteristics N %

Years of experience with B2B purchasing (Variable: Experience) 1-2 years 3-5 years 6-15 years 16-25 years More than 25 69 106 75 18 10 25 38 27 6 4 Amount of time spend on purchasing tasks (Variable: Purchasing_time)

Full time (36 hours or more) Part time (Less than 36 hours) Now and then

147 76 58 53 26 21 Number of employees of the company of the participant (at that specific

location) (Variable: Nr_employees) 0-9 10-24 25-49 50-99 100-249 More than 250 43 69 47 23 35 61 15 25 17 8 13 22 Purchasing department context (Variable: Separate_dpt)

No separate purchasing department

Separate purchasing department, but not employed there Separate purchasing department, employed there

28

The participants were randomly assigned to one of the four vignettes. The two low uncertainty groups together (vignette 1 and 2) contained 143 participants. Of which 73 participants were in the positive framed group (vignette 1) and 70 participants were in the negative framed group (vignette 2). The two high uncertainty groups together (vignette 3 and 4) contained 135 participants. Of which 66 participants were in the positive framed group (vignette 3) and 69 participants were in the negative framed group (vignette 4). The characteristics are displayed in figure 3.

Figure 3 - Boxplots

The deletion of the far-outliers will increase the mean and decrease the variance of the supplier evaluation scores for vignette 2. The descriptive statistics without data point 14, 81 and 86 are listed in table 4.

Table 4 - Descriptive statistics without far-outliers

Framing Uncertainty N Min. Max. Mean Standard Deviation

Variance

29 Manipulation check

To measure how participants perceived uncertainty, two variables (derived from Milliken (1987) and Riedl et al. (2013)) were included in our research. The variables were measured with statement Q4 “I had too little information to rate supplier.” and statement Q5 “I felt able to predict accurately what the consequences of my rating might be, for the <focal firm of vignette>.”.

As both the high and low uncertainty groups are relatively large, therefore we assume normality in each group (Lumley et al., 2002). No significant difference between the high and low uncertainty groups was found for statement Q4. For statement Q5 there was a significant difference in means between the high and low uncertainty groups. However, we cannot assume equality of variances because the Levene’s test shows a significance value of 0.005, which is smaller than 0.05.

Therefore, a non-parametric Mann-Whitney test is conducted to compare the scores of both groups on statement Q5. For the Mann-Whitney test to be significant, the U needs to be less than a half times the product of the sample size of the high uncertainty group (N=135) times the sample size of the low uncertainty group (N=143), that is lower than 9652,5. The p-value should be lower than 0.05. The Mann-Whitney test shows a U of 7523.5 and a p-value of 0.001. The test further shows a correlation coefficient of 0.204.

Thus, the low uncertainty group (N=143, M=3.69, SD=3.25) scored significantly higher (P=0.001) on this statement compared to the high uncertainty group (N=135, M=3.25, SD=1.08). This indicates that the participants perceive significantly different degrees of uncertainty between the uncertainty groups.

Hypothesis 1

Hypothesis 1 was stated as follows: ‘Supplier selection professionals are susceptible to the framing bias.’ To test hypothesis 1, the positively and negatively framed groups have to be analysed. First, we will analyse the normality of the supplier evaluation scores within each group. Second, we will analyse and compare both framed groups with low uncertainty (vignette 1 and 2). Third, we will analyse and compare both framed groups with high uncertainty (vignette 3 and 4).

30

assumed for group 2 (p=0.98) and 3 (p=0.170). Group 4 has a small violation of the assumption of normality (p=0.048).

Group 1 shows a relatively large violation of the normality assumption. Additionally, a Levene’s test has been used to check if we can assume equal variances between group 1 and 2. We can assume equal variances between groups if the significance of the Levene’s test is larger than 0.05. The Levene’s test shows a value of 0.014. This means we cannot assume equal variances. Therefore, we used a Mann-Whitney test as a non-parametric alternative for the t-test. When there are significant differences between the two groups, the U-value should be smaller than half of the product of the group size of group 1 times the group size of group 2 (here <2555) and the p-value should be smaller than 0.05. The Mann-Whitney test (U=938.0, p=0.000, r=0.548) shows that the differently framed groups have significant different means with a strong correlation coefficient (r) of 0.0548. The average supplier evaluation score of the positively framed group 1 (N=73, M=81.63, SD=9.61) is significantly higher than the average supplier evaluation score of negatively framed group 2 (N=70, M=67.04, SD=13.91). These results support hypothesis 1.

An independent-samples t-test was conducted to compare the average supplier evaluation scores of group 3 and 4. For group 3 and 4 we can assume equal variances (The Levene’s test shows a significance value of 0.866) and there was only a relatively small assumption of normality for group 4. The t-test can be used to compare both groups, as the t-test only requires approximately normal data. The t-test is quite robust to violations of normality (Laerd Statistics, 2017). Vignette 3 is positively framed (N=66, M=77.86, SD=12.06) and vignette 4 is negatively framed (N=69, M=69.33, SD=12.76). Also between these groups, the results shows that both groups have significant different means (P=0.000) of the supplier evaluation scores. These results also support hypothesis 1. From both the independent-samples t-tests and the Mann-Whitney test we can conclude that hypothesis 1 can be confirmed (the statistical tests are presented in appendix I).

Hypothesis 2

31

independent-samples t-tests. The Levene’s test shows a significance value of 0.07, which is larger than 0.05. This means we can assume equal variances.

After looking at the assumptions, a two-way ANOVA was conducted to test hypothesis 2. First, all participants were divided into two groups (high and low uncertainty). The first group (N=143, M=74.49, SD=13.94) consists of vignette 1 and 2 (low uncertainty) and the second group(N=135, M=73.50, SD=13.09) consists of vignette 3 and 4 (high uncertainty). The relationship between uncertainty and the supplier evaluation score is not significant (P=0.613). Although, we know that the framing effect seems to be stronger for the low uncertainty groups, the means of the high and low uncertainty groups together do not differ significantly.

Second, all participants were again divided into two groups (positive and negative framing). The first group (N=139, M=79.84, SD=10.97) consists of vignette 1 and 3 (positive framing) and the second group (N=139, M=68.18, SD=13.35) consists of vignette 2 and 4 (negative framing). The relationship between framing and the supplier evaluation score is significant (P=0.000).

32 Figure 4 – Moderating effect of uncertainty

Additional Statements

Each participant had to state their agreeableness to fourteen different statements on a five-point Likert scale (from strongly disagree to strongly agree). We have investigated for each statement if the means of the scores significantly differ between the high (vignette 3 and 4) and low (vignette 1 and 2) uncertainty groups. The statements do not measure what the effects are of the framing effect and uncertainty (as a moderator) on the supplier evaluation score. Instead, the statements try to predict why uncertainty moderates the framing effect.

33

with ‘**’. In table 5 we can see that there are seven statements that significantly differ between the high and low uncertainty groups (see also appendix IV).

Table 5 - Post hoc analysis (statements)

*=equality of variances assumed, **=significant difference between groups.

Statement Group 1, N=143 (low uncertainty) Group 2, N=135 (high uncertainty) Levene’s test IS T-test P-value M SD M SD

Q4. I had too little information to rate the supplier.

3.29 1.12 3.31 1.01 0.151* 0.892

Q5. I felt able to predict accurately what the consequences of my rating might be, for the <focal firm of vignette>.

3.69 0.90 3.25 1.08 0.005 0.000**

Q6. I tried to take as much information into account as possible to make a good decision.

4.13 0.77 3.83 0.89 0.279* 0.03**

Q7. I tried hard to make an optimal decision.

4.03 0.86 3.81 0.89 0.306* 0.036**

Q8. I found it hard to rate supplier X’s performance.

3.29 1.13 3.15 1.03 0.185* 0.286

Q9. In my professional career, I had to make purchasing decisions before.

3.95 0.93 3.87 1.04 0.339* 0.475

Q10. I know something about the framing effect (a cognitive bias).

3.41 1.12 3.25 1.14 0.823* 0.237

Q11. I think the task to rate supplier X was mentally demanding.

3.15 1.16 3.10 1.02 0.095* 0.742

Q12. I have a strong personal opinion about <industry of vignette>.

3.11 1.23 2.78 1.12 0.339* 0.019**

Q13. I am personally involved with <industry of vignette>..

2.81 1.36 2.60 1.39 0.436* 0.203

Q14. I used Calculations/statistics to rate supplier X.

3.26 1.23 2.96 1.18 0.516* 0.037**

Q15. I used professional purchasing logic/rules/matrixes/schemes to rate supplier X

3.25 1.17 2.91 1.18 0.941* 0.016**

Q16. I used rational reasoning to rate supplier X

3.97 0.87 3.71 0.84 0.329* 0.014**

Q17. I used my feelings/intuition to rate supplier X.

34

After answering the fourteen statements, each participant had to rate seven text fragments from the vignette on their importance to rate supplier X. Each text fragment was rated on a five-point Likert scale (Not important, slightly important, moderately important, important, very important). In table 6, the results are listed. Four text fragments were found to score significantly higher in importance in the low uncertainty groups (marked with ‘**’). For these significant differences, we can assume equality of variances. Amongst these text fragments were also the framed text fragments TF4 and TF5 (see also appendix IV).

Table 6 - Post hoc analysis (text fragments)

*=equality of variances assumed, **=significant difference between groups. Post Hoc

After testing our posed hypotheses statistically, we can conclude that the framing of supplier characteristics and uncertainty_framing (as a moderating variable) can significantly affect supplier evaluation scores.

In the post hoc analysis, a multiple linear regression was used for our control measures (framing, uncertainty and uncertainty_framing) to assess the ability of each variable to predict supplier evaluation scores, after controlling for the influence of control variables (age, gender and duration) and supplier selection related variables (Experience, Nr_employees, Purchasing_time and Separate_ dpt). Additionally, we have investigated if other supplier selection related variables are significant predictors of variance in the supplier evaluation score.

Text fragment Group 1, N=143

(low uncertainty) Group 2, N=135 (high uncertainty) Levene’s test IS T-test P-value M SD M SD TF1.Supplier X is internationally operating. 3.77 1.00 3.68 0.93 0.617* 0.451

TF2. Supplier X uses high quality products.

4.41 0.77 3.99 0.88 0.680* 0.000**

TF3. Supplier X delivers slightly below market price.

4.02 0.90 3.8 0.94 0.241* 0.046**

TF4. Supplier X delivers <X out of Y on time/too late>.

4.20 0.91 3.87 1.02 0.183* 0.004**

TF5. Supplier X delivers X <% (in)sufficient quality>.

4.25 0.94 3.99 0.93 0.212* 0.018**

TF6. The quality ratings of other colleagues (70 and 74%).

3.71 1.04 3.48 1.01 0.527* 0.069

TF7. Supplier X is on the market for X years.

35

First, a multiple linear regression is used to investigate to what extent framing, uncertainty, and uncertainty_framing (as a moderating variable) can predict the supplier evaluation score in our vignette-based experiment. This type of regression allows for correction of other variables. In this way, we can correct for the influence of control variables and supplier selection related variables.

In the regression analyses we found that framing, uncertainty and uncertainty_framing (moderating variable) are all significant predictors of the supplier evaluation score. By adding the moderating variable to the regression, the interpretation of framing and uncertainty changes. In this way, framing and uncertainty cannot be only interpreted as self-contained predictors of the supplier evaluation scores. The p-values of each predictor is respectively 0.00, 0.032 and 0.041. The Pearson correlations of the variables are respectively -0.432, -0.037 and -0.227. From this, we can conclude that framing and uncertainty_framing are strong significant predictors of the supplier evaluation score variances. Uncertainty is also a significant predictor of the supplier evaluation score variances. Although, uncertainty can only explain a small part of variance in the supplier evaluation scores, and therefore has relatively little predictive power. The model with these three variables together is found to be significant. For this model, F (10.267) = 8.916 and p = 0.000 (for a more detailed analysis see appendix V).

Second, this post hoc analysis investigates if control or supplier related variables significantly affect the supplier evaluation scores. To validate our research we need to investigate if there are other variables, besides from our control measures, that affect the supplier evaluation scores. Possible findings can also be used as directions for further research.

36

Purchasing_time is a significant (p=0.035) predictor of the supplier evaluation score in vignette group 1 with a correlation coefficient of 0.260 (see also appendix V).

Discussion

Carter et al. (2007) suggest an examination of the effect of biases in decision-making in supply chain management situations under different degrees of uncertainty. As a whole, this research tries to fill the gap between the theorized influence and actual influence of biases on decision-making in supplier selection under uncertainty. This research contributes to the field of behavioural supply chain management by examining the framing bias in supplier selection under different degrees of uncertainty. Thereby, this research offers new perspectives on how human behaviour (in the form of the framing bias) can influence decision-making in supplier selection.

The framing effect

Our first posed hypothesis is “Supplier selection professionals are susceptible to the framing bias.” The Mann-Whitney test showed significant differences in supplier evaluation scores between low uncertainty groups. In the low uncertainty context, participants rated supplier X on average 81.63% with positively framed supplier characteristics. In the same context, with negatively framed supplier characteristics the average supplier evaluation score was 67.04%. As only the framed supplier characteristics have been manipulated between both groups, we can conclude that the framing of supplier characteristics led to a significant difference of 14.59% in average supplier evaluation scores.

The two groups in the high uncertainty context were compared with an independent samples t-test. Here, the group with positively framed supplier characteristics rated supplier X with an average of 77.86%. The group with negatively framed supplier characteristics rated supplier X with an average of 69.33%. The difference of 8.53% between both groups is smaller compared to the two groups in the low uncertainty context, but also significant. Therefore, our first hypothesis is accepted.

37

Role of uncertainty

Our second posed hypothesis is: The degree of uncertainty in supplier selection moderates the framing effect, such that the framing effect will be stronger as uncertainty increases. Our research has provided new insights about the role of uncertainty in supplier selection decisions. As already stated in the introduction, Carter et al. (2007) have suggested an evaluation of uncertainty as a moderating effect on decision biases in supply chain management decisions as a direction for further research. This research contributed to this area by focusing specifically on the framing bias.

The statistical tests conducted to test hypothesis 1 already showed that the means of the supplier evaluation scores were further apart for the low uncertainty group compared to the high uncertainty group. A two-way ANOVA was conducted to investigate the moderating effect of uncertainty. The results of this test showed that the moderating effect of uncertainty was significant (see Appendix II). However, we have to reject the second hypothesis. The results of the two-way ANOVA showed the opposite of what we expected.

These findings contribute to traditional supplier selection models. The classification of the context in supply chains decisions is often used as a starting point for supplier selection decision-making. Logically, the rational supplier assessment approach will have different outcomes if the supplier was located in a different context with different uncertainties. However, our research shows that uncertainty does not only influences rational decision models, but also influences the way supplier attributes are interpreted when the information is framed.

38 Figure 5 – The framing bias moderated by uncertainty

This contradicts multiple findings in the field of cognitive science (Kahneman, 2012; Tversky and Kahneman, 1974) . Also from a behavioural supply management perspective, we would expect that decision-makers are more vulnerable to biases under uncertainty (Carter et al., 2007).

However, these findings do contribute to the works of Kaufmann et al. (2009) and Lipshitz and Strauss (1997) who state that decision-makers will actively try to reduce the uncertainty. As a result of this strategy, the framing effect could have less effect under high uncertainty. The framing effect could be reduced as a result of the information gathering tactic, used for reducing uncertainty (Kaufmann et al., 2009; Lipshitz and Strauss, 1997). A participant using this tactic, tries to make an optimal and well-informed decision by taking all information into account when confronted with uncertainty. As the framed pieces of information only represent a relatively small part of all the available information, the framed information could hold less sway in the decision-making.

39

possibilities to use this tactic. This tactic has similarities with Tversky and Kahneman's (1974) ‘adjustment from the starting point’ heuristic. As the adjustment from the starting point is typically insufficient (Tversky and Kahneman, 1974), the heuristic fails to produce a correct judgement which will result in the anchoring bias (Botros et al., 2014; Kahneman, 2012). If we look at the ratings of colleagues as anchors, the ratings of colleagues hold significant sway over decision-making, in such a way that the rating of colleagues is overqualified and the other available information is underqualified (Botros et al., 2014). It is therefore likely, that the measured supplier evaluation scores under different degrees of uncertainty are the result of an interplay of the framing bias, and the anchoring bias. As biases particularly occur under uncertainty, the supplier evaluation scores are more strongly ‘anchored’ to the ratings of the colleagues in the high uncertainty situation. In this way, our research further extends the previous works of Carter et al. (2007) on decision-making biases in supply chains. Our research shows that multiple biases can have opposite effects on supplier evaluation scores in such decision-making processes. The interplay between the framing and anchoring bias is visualized in the figure below. In this figure, the blue arrows visualize the anchor effects under both high and low uncertainty and the blue dotted line visualizes the anchor caused by the colleagues’ advice in the vignette.

40

The understanding of decision-making biases and the interplay between these biases can be the beginning of eliminating biased decision-making in supply chains. Previous research of Kaufmann et al. (2009) already showed that there are many debiasing strategies available to counter individual biases. However, to debias decision-making, researchers must understand biases itself. Where Kaufmann et al. (2009) only focus on individual biases, our research shows that interplays of biases can also be possible. In this way, researchers are able to better understand biased decision-making in supplier selection decisions under different degrees of uncertainty. In addition, researchers are able to better debias decision-making if they acknowledge and understand the interplay between multiple biases in decision-making.

Finally, our findings build on previous studies about supplier selection behaviour under uncertainty of Kull et al. (2014) by investigating determinants of decision-making under different degrees of uncertainty. In this way, our research contradicts the traditional supply chain management view. In particular, our research underlines that complementary to previous beliefs, behavioural aspects help explain how and why supplier selection professionals make certain decisions.

Additional statements

The vignette-based experiment already showed that the different experimental groups had significantly different supplier evaluation scores. As described in our methodology section, statements were added to explain why the results between groups might be different. According to literature, there can be multiple explanations for our findings regarding the moderating effect of uncertainty on the framing effect. Our statements were based on these possible explanations. The average scores (on a five-point Likert scale) were compared between the high and low uncertainty conditions for each statement by conducting an independent samples t-test for each statement. No evidence could be found that the stronger framing effect in the low uncertainty groups was caused due to participant’s reduced cognitive efforts (Koufteros et al., 2002), intuitive way of reasoning or relying on heuristics (Pitz and Sachs, 1984).