Can trust induce vertical integration? An experimental study of buyer–seller exchanges with distinct competencies and specific investments

Abstract

1. Introduction

A central question in strategy research revolves around whether to organize activities in markets (i.e. outsourcing) or hierarchies (i.e. in the firm), with transaction cost economics (TCE) having emerged as a widely articulated theory to grasp such choices. Based on initial ideas by Coase (1937) and Williamson (1985, 1991) explains TCE’s fundamental premise as that of firms choosing vertical boundaries based on tradeoffs between organizational costs and market failure; so increasingly hazardous market exchanges induce firms to vertically integrate, while low-risk ones induce them to outsource instead. These hazards, Klein et al. (1978) explain, largely involve transaction-specific assets which, since they are worth much less in alternative deals, incite opportunistic parties to hold up or renege on previously agreed promises. At a base level, the choice to vertically integrate helps avert such difficulties, as firms can allocate gains by mandate; although when firms have reasons to trust—that is, to believe in the trade partner’s reliability, benevolence, and integrity (Zaheer et al., 1998)—they will opt to outsource instead, since trust mitigates various sorts of market failure and fosters exchange continuity (see Poppo et al., 2008).

Although not denying this view, in this article, we propose a distinct channel through which trust can instead trigger more, not less, vertical integration. This alternative hypothesis emerges when, besides the usual market failure arguments, we consider the distinct competencies of transacting parties. Prior studies point that firms develop distinctive competencies by specializing in their own craft along the value chain; so, following the logic of comparative advantages (e.g. Langlois and Robertson, 1995; Jacobides, 2008; Colfer and Baldwin, 2016), exchanges can be organized in markets as firms seek to build not only alternative sources of knowledge but also opportunities for trade gains (e.g. Jacobides and Hitt, 2005). However, parties may choose to combine their distinctive competencies in a vertically integrated firm when internal production gains become available from tighter and closer coordination (Argyres and Zenger, 2012; see also Chandler, 1990; Kogut and Zander, 1992; Teece, 2010; Helfat, 2015). So what holds firms from seeking multiple such prospects for vertical integration?

The answer is that hierarchies are known to bring the downside of higher internal bureaucratic costs. Without the high-powered incentives of markets, hierarchies can be plagued with agency costs, as internal agents may shirk and reduce productivity-enhancing efforts that may be best brought about in market transactions (e.g. Gibbons, 2005, Puranam et al., 2006, Zenger et al., 2011). We believe trust can have an agency mitigating role, and facilitate vertical integration. For example, GE Aviation’s Vice President was inquired about the acquisition of Morris Technologies in 2012, largely as a function of its valuable complex geometry design and precision manufacturing competencies. He justified GE’s choice by pointing to the historically trusting and close relationship between the two firms as a significant factor (Aviation, 2012). In other words, trust may not directly cause but perhaps facilitate vertical integration as it can increase the perceived value of pooling distinct, specialized competencies. Though usually seen as a leading contributor of transaction cost-based theories of the firm, Coase (2000, 2006) himself has called for an encompassing view of the determinants of firm boundaries. In this view, the decision to acquire or integrate with an exchange partner is crucially influenced by a history of interactions, leading to a desire to strengthen production and capitalize on their trust-based relation (Coase, 2000, pp. 26–27).

This is an intriguing hypothesis generating a different prediction on the facilitating role of trust in vertical integration decisions, though one that unfortunately has not been fully developed conceptually, nor further investigated empirically. A particular difficulty is that empirically, in the real world, it is hard to tease out competencies that are distinct (and so hold the promise of complementarity) from those that are exchange specific. Our experimental methodology addresses this issue. Using structured interactions in a laboratory setting, we pair up 806 buyers and sellers, and have them choose markets or hierarchies under treatments that combine different levels of specific investments, distinct competencies, and trust. Based on the concept of learning curves, we manipulate distinct competencies by allowing sellers to practice the assembly of a Lego^® car, with production costs being a direct function of assembly time. Given that trust knowingly reduces opportunism, we argue it can mitigate internal (agency) hazards, so the likelihood of vertical integration to internalize distinct competencies increases if partners are perceived to be trustworthy. Running two complementary experiments varying exchange conditions influencing competencies, specific investments, and trust, we find strong support for the hypothesis that trust facilitates vertical integration in a context of distinct competencies.

Our study aims to address an evolving debate on the role of trust in organizational choices. In short, we advance the scholarly debate that whether firms vertically integrate likely depends on a combined assessment of both transaction costs and capability-based determinants (Langlois and Robertson, 1995; Jacobides, 2008; Jacobides and Winter, 2012). Specially, we show that trust not only promotes markets by reducing the hazards of specific investments (as emphasized by previous work), but alternatively triggers vertical integration when parties would like to internalize distinctive competencies and at the same time avoid hierarchical failure. Empirically, our experimental design not only improves the assessment of causality, but also implements a controlled exchange setting, with standardized tasks that allow us to uniformly gauge the effect of distinct competencies and separate them from other sources of exchange heterogeneity as well as myriad confounds.

2. Theory background

TCE’s case for vertical integration is well articulated. Exchange hazards emerge in the classic holdup problem, where one party makes transaction-specific asset investments that (because they are worth less in alternative deals; Klein et al., 1978) are subject to a risk of opportunistic ex post exploitation by the partner (who reneges agreed terms; e.g. Williamson, 1985). Given partner behaviors are uncertain and contracts are naturally incomplete, firms vertically integrate to avoid this risk (Klein et al., 1978). A key boundary condition in this scenario is that if contexts involve trust, contractual hazards naturally decrease (Williamson, 1991), therefore leading firms to choose markets. Hence, trust negatively moderates the effect of specific investments on vertical integration, because it lowers holdup risks (Gulati and Nickerson, 2008) and promotes exchange continuity (Williamson, 2008).

Following foundational work by Penrose (1959) and Nelson and Winter (1982), scholars have also examined vertical boundaries based on the logic of heterogeneous competencies (e.g. Jacobides and Hitt, 2005). In contrast to transaction cost logic, the competencies reasoning justifies the choice to internalize an activity based on the relative gains to access productive knowledge within the firm versus outside. The concept of distinct competencies reflects the notion that a firm’s resources are superior to those of others, even if such competencies are tradable with many other partners (i.e. non-exchange specific; Gibbons, 2005). Hence, firms with better production competencies in one supply chain stage tend to perform that activity internally (Jacobides and Hitt, 2005: p. 1209) and scan the market to contract with another where they are deficient. In this view, markets emerge as a way to promote exchange gains among specialized agents holding distinct (and complementary) competencies (Langlois and Robertson, 1995; Jacobides, 2008; Colfer and Baldwin, 2016).

But recent organization studies instead suggest that internalizing (rather than outsourcing) distinct competencies may also offer the potential for tighter know-how integration and coordination. For instance, parties may come to share a common identity and language (e.g. Kogut and Zander, 1992; Cremer et al., 2007); manage multiple, synergistic activities subject to scope economies (Chandler, 1990; Teece, 2010); or build systemic platforms exploiting the complementarities of distinct capabilities in the value chain (Helfat, 2015). In these conditions, suboptimal outcomes may materialize in the absence of tight joint problem solving and mutual adaptation (Jacobides and Winter, 2012). Gibbons (2005: p. 19) in fact explains that this alternate view provides “a coherent elemental theory of the firm without specific investments,” describing make or buy choices as a function of enhanced coordination in hierarchies instead of holdup and other potential sources of market hazards.

The internalization of distinct competencies however is not an axiomatic choice. By vertically integrating a supplier, a buyer is subject to potential agency costs; that is, in a hierarchy, low-powered incentives can attenuate the synergic gains of the internalized competencies (Argyres and Zenger, 2012). Agentic issues arise not only due to the complexity of the integration, but from autonomy incentives as well. For instance, one’s technical competencies may involve know-how that is difficult to measure and explain, and so need time-consuming alignment (Puranam et al., 2006). All the while, firms must fine-tune the ideal internal autonomy of operations. Typically, sellers in outsourcing relations have extensive autonomy, but end up frustrated as their discretion shortens when brought in house. Therefore, agency-related difficulties can quickly eclipse internal synergies, especially in situations when the expertise of the two firms is highly distinct and specialized. In other words, as opposed to the usual TCE focus on market failure, internal sources of hierarchical failure may undermine the integration of distinct competencies.

It is here that we advance our theoretical argument that trust helps attenuate this risk of hierarchical failure. Primarily, trust reflects a situation where one is willing to be vulnerable to the acts of another when there is a non-trivial probability of loss due to deliberate cheating or unrealized expectations (Mesquita, 2007). Thus, a transacting party will form an expectation that the other partner will act in a trustworthy manner. Research has shown that these expectations are largely influenced by social conditions surrounding the exchange. One may trust simply out of identifying herself with the goals of another, even if such impressions are based on impromptu communications that promote the alignment of goals and expectations. In this latter case, game theorists argue that pre-play interaction allows parties to exchange information and arrive at more beneficial equilibrium outcomes (e.g. Crawford, 1998; Farrell, 1988; Ellingsen and Johannesson, 2004). Parties that many times are mutual strangers can apply informal heuristics such as reciprocity, however flimsy they may be, and evaluate cues on how close their expected gains and personal goals are, coming to perceive the other as reliable or not. Alternatively, trust may emerge from a previous history of interactions, as it happens in relationally embedded contexts where parties have existing social attachments (e.g. Granovetter, 1985; Ruef, 2002), and develop mutual social norms of behavior after being acquaintances for a time. Such notions of trust capture ideals of goodwill, honesty, fairness, openness, caring intentions, as well as predictability (Colquitt et al., 2007: p. 910).

To sum up, we model the seemingly opposite roles of trust as a boundary condition in make-or-buy decisions. In contrast to the logic based on the threat of holdup—that trust negatively moderates the vertical integration of specific investments—our alternate hypothesis poses that trust will instead positively moderate the vertical integration of distinct competencies (Figure 1). To enhance the theoretical precision of our reasoning and provide a more direct connection with our experiments, below we further develop a game-theoretic negotiation model that helps us formally enunciate our hypotheses.

Figure 1.

Model summary.

Open in new tabDownload slide

3. Formal model and hypotheses

Let us consider a buyer and a seller (herein “she” and “he,” respectively). We start with the buyer seeking to explore a market opportunity for a given good (in our case, vehicles), so it must decide whether to make or buy it. To get the vehicles produced, she tries to convince a seller to invest in a new plant (which reflects the production technology and capacity needed) by promising to buy the output produced. If the seller makes the investment, buyer and seller have a market arrangement. Though it is customary for partners to write a contract at this stage to specify mutual pledges, contracts are incomplete so the buyer may ex post renege and demand renegotiation of the price for which the seller would manufacture and deliver the good (Hart and Moore, 1988). Wary of this risk, the seller may refute the initial proposal. The buyer in this case can offer another deal where the seller still builds the plant but immediately sells it to the buyer, while agreeing to assemble the car, for a fixed fee, as an employee in the buyer’s firm. This is equivalent to an acquisition with job hiring, in which the other’s know-how is vertically integrated, so we shall refer to it as a vertical integration with hiring (i.e. “with distinct competencies”); that is, the buyer pays for the know-how of the seller and internalizes production as well as the distinct competencies of the seller. This arrangement thus mirrors an internal employment relation, with a previous independent seller hired as an employee of the buyer (e.g. Lafontaine and Slade, 2007). In case both proposals fall apart, we give the buyer a third (fallback) option, in that she can still invest in the plant and produce the good all by herself; a mode which we shall refer to as vertical integration with own production (i.e. “without distinct competencies”). In this latter option though, the buyer fails to benefit from the distinct competencies of the seller, and as a single agent has to perform all activities in the production chain. It is obvious that our core theoretical interest lies with the hiring option, for it examines the interaction of trust and distinct competencies, but the inclusion of the own production fallback option eases trade tensions so ensures the hiring choice is not incurred due to a lack of alternatives.

Formally, the buyer can purchase the good produced by the seller at a price p and then resell it for a value of v > 0 (net of all selling costs). The production requires an up-front plant investment of I > 0, which the buyer wishes the seller will agree to incur. In case the seller refuses the initial proposal of p, the buyer must carry out the investment, which for simplicity we assume it is also I, in all cases: market or vertical integration (with hiring or own production). Buyer and seller have respective production costs of c_B and c_S, where, if the seller is specialized in the production of the good, |${c_B} \gt {c_S}$|⁠. In fact, |$({c_B} - {c_S})$| indicates the extent of distinct competencies between buyer and seller. For simplicity, we assume that |${c_B}\,\,{\rm{and}}\,\,{c_S}$| are common knowledge, so both parties know their relative efficiencies in producing the good.

3.1 Market arrangement

A market arrangement occurs if the seller opts to make the investment for a given pre-specified supply price p (so I is automatically debited from the seller’s profit, in our game-theoretic framework). In agreement with Hart and Moore’s (1988) idea of incomplete contracts, buyers and sellers can renegotiate p in the subsequent period. Thus, in the next period, buyer and seller profits will be respectively |$v-p$| and |$p - {c_S} - I$|⁠. We assume |$v-{c_S}-I \gt 0$|⁠, so a positive surplus can be divided between them. If they fail to reach an agreement on p though, the buyer receives zero and the seller pays for the non-redeployable (sunk) part of the investment, |$k(0 \le k \le 1)$|⁠, thereby incurring a loss of – kI. In this case, parameter k represents a measure of specific investment; if k = 1, then the investment is totally non-redeployable. It is instructive to note the difference between investment specificity and distinct competencies in our context: even when sellers are not required to customize their production process to a specific buyer (i.e. k = 0), they may still be highly skilled (e.g. serving the broader market) and hence be much more efficient than the buyer in the production of the underlying good (i.e. c_S can be much lower than c_B).

In this setting, if specific investments are required, holdup will potentially occur precisely due to the sunk portion of the initial investment. Given a certain price offer from the buyer, in the second period the seller receives |$p-{c_S}-I$| if he agrees with the new price p or – kI if he refuses it. So, the seller viably accepts any renegotiated price |$p \geqslant {c_S} + I\left( {1 - k} \right)$|⁠. Opportunistic buyers can thus force a negotiation over the ex post surplus |$p \ge {c_S} + I(1 -k)$|⁠. Assuming a Nash bargaining solution where they equally split the surplus, buyer and seller finally agree on a price |$p = [v + {c_S} + I(1 -k)]/2$|⁠. The seller ex ante profit then equals |$p-{c_S}-I = [v-{c_S}-I(1 + k)]/2$|⁠.

3.2 Vertical integration

In contrast to the market arrangement situation, vertical integration occurs if the seller refrains from investing in the plant, so the buyer herself must invest. In this case, the pair can still negotiate a possible transfer of know-how by the seller whenever |${c_B} \gt {c_S}$|⁠. In line with our prior literature review, we assume that vertical integration of distinct competencies facilitates the coordination of interdependent production processes in the value chain. Although for simplicity we do not model these synergies directly, we suppose that in the mode of vertical integration, production costs drop by a factor |$\delta 0 \leqslant \delta \lt 1$|⁠). In this case, costs become |$(1-\delta ){c_S}$| if the seller does the assembling (i.e., vertical integration with hiring) and instead |$(1-\delta ){c_B}$| in the own production form of vertical integration. The production synergies from vertically integrating the seller’s distinct competencies occur through a hiring agreement where the buyer acquires the seller, hires him to work for a fee x, and incurs the costs of production (i.e., vertical integration with hiring). If they fail to reach an agreement on x, the seller earns zero, while the buyer engages in the production process herself, thus receiving v–(1–δ)c_B–I (i.e., vertical integration with own production). We assume that |$v- (1-\delta ){c_B}-I \ge 0$|⁠, so it is still profitable for the buyer to invest in the plant and produce the good all by herself, even considering her higher production cost c_B.

3.3 Determinants of vertical integration

If equation (3) holds, then vertical integration with hiring is expected, being that the higher the left-hand side of inequality (3), the higher the likelihood of this vertical integration mode. We can thus immediately check that vertical integration will be discouraged if the term |$[v - {c_S} - (1 + k)I]$| is large, indicating that there is a high potential value to be captured in the market arrangement. Notice that this term increases (thus making vertical integration less attractive) if the seller’s cost (c_S) decreases, which is aligned with the common view that the presence of more efficient or capable partners encourages outsourcing (Jacobides and Hitt, 2005). Our key terms of interest however are not in the direct effects, but in how trust moderates the make-or-buy choice, or |${\alpha _S}(1 - \delta ){\mkern 1mu} ({c_B} - {c_S})\;{\text{and}}\; - {\alpha _B}kI$|⁠. The latter term formalizes the known prediction that, with specific investments (i.e. high k), an increase in trust—a growing belief that the buyer will not act opportunistically (i.e. higher α_B)—reduces the value of vertical integration and promotes markets (Williamson, 1991). Notice, however, that this effect depends on trust (α_B). In inequality (3), the total effect of specific investments is |$kI - {\alpha _B}kI$|⁠. Thus, in a low trust environment (α_B very small), the last term approaches zero and we should expect the familiar prediction that specific investments should promote vertical integration. In contrast, in a high trust environment (with a high α_B), the effect of specific investments on integration should diminish. Also, note that the effect of α_B in reducing the value of vertical integration is more pronounced when k increases (i.e. with relevant specific investments to be made, trust helps overcome contractual hazards).

In turn, the former term |${\alpha _S}(1 - \delta )\;({c_B} - {c_S})$| formally encapsulates our alternate hypothesis that, under distinct competencies, or a high |$({c_B} - {c_S})$|⁠, higher trust (i.e. an expectation that sellers will not act opportunistically; that is, high α_S) leads to an increase in vertical integration with hiring. Again, the effect is highly dependent on trust; the value of distinct capabilities in a context of vertical integration depends on the expectation that the seller will perform after he is hired in the integrated firm. Said differently, trust increases the confidence the buyer can hire the seller, avoiding agentic opportunism when competencies are internalized. Formally, we state these effects as follows:

H1: An increase in trust (i.e. a rise in the seller’s perceptions that the buyer is trustworthy) reduces the effect of specific investments on the likelihood of vertical integration (with hiring).

H2: An increase in trust (i.e. a rise in the buyer’s perceptions that the seller is trustworthy) enhances the effect of distinct competencies on the likelihood of vertical integration (with hiring).

4. Experimental methodology

To test our predictions, we conduct two controlled laboratory experiments involving vertical integration choices—the first of which follows the interaction structure of the previous section and tests both H1 and H2, while the second changes the sequence of decisions and specification of forms of trust to more specifically scrutinize the effect predicted in H2, while abstracting from up-front-specific investments and introducing the role of production synergies (via parameter δ).

A crucial contribution of our design is to separate the effect of specific investments from the effect of distinct competencies, and how these conditions are moderated by trust. Given our controlled setting, we are able to uniformly vary and gauge distinct competencies. Namely, we adopt an assembly task by means of a 41-piece Lego^® toy (a Lego^® vehicle, model no. 8122 “Desert Viper,” henceforth “Lego^®” or “vehicle.” Supplementary Appendix SA presents an illustration of a participant assembling the Lego^® toy). The assembly requires both know-how and investment in a (fictitious) plant. The know-how to assemble a Lego^® is subject to learning curve effects, such that the usual assembly time drops in proportion to one’s sequential practices (in our study, 40% after the second time, 60% after the third, 70% after the fourth time, and so on). We accordingly exploit this fact as a source of distinct competencies, that is, in some sessions, sellers but not buyers practice the assembly. We explicitly show all participants a graph with these statistics to make them aware of the learning curve, so players can judge each other’s competence advantages. In each negotiation session, we compute players’ costs and gains at each phase with experimental points, and only at the end convert these into cash payments. We time the Lego^® assembly done by our experiment subjects, given that production costs are a function of time.

Early in each session, we read instructions and quiz participants to assure understanding, such that responses were individually checked and corrected to guarantee sufficient knowledge of experiment procedures. Finally, we randomly assigned subjects to buyer and seller roles. We paid subjects cash, including a $5-dollar show-up fee, plus a variable premium based on performance. Average pay was around $32 dollars. Sessions typically lasted one hour. More details on the comparison between the experiments, experimental protocol in each case, and all survey instruments are included in the Supplementary Appendices SB–E).

5. Experiment 1

5.1 Experimental setting

The sequence of decisions in this experiment is directly connected with the former theory model. Thus, we simulate a buyer–seller negotiation where the buyer is said to be an automobile company with a market opportunity for a novel vehicle. By selling this vehicle, the buyer will earn gross revenues of 100 experimental points (i.e. v = 100 in the model presented earlier). To get the vehicles produced, the buyer wishes to convince a seller to invest in a new plant, so the seller would supply the completely assembled car to the buyer, at a supply price p that is mutually negotiated. Such plant investment costs 40 points (i.e. I = 40). If the seller refuses to invest in the plant, the buyer must make the investment herself, for the same 40 points; in this case, the seller agrees to build the plant and the buyer immediately acquires it, so the seller, now as a hired employee, assembles the car in the buyer’s factory. If the seller refuses both offers, then the buyer can still invest in the plant (I = 40) and assemble the car herself. In line with our formal negotiation model, the experiment evolves in three sequential phases (Figure 2).

Figure 2.

Experiment 1: sequential phases.

Open in new tabDownload slide

5.1.1 Phase 1: seller’s decision to invest or not in the plant

In this phase, the seller decides whether to invest in the plant. Should the seller agree to make the specific investment, he will incur I = 40 points, as explained above. In this case, buyer and seller will go to phase 2a, whereby they can renegotiate the final supply price p which the buyer is to pay the supplier for the assembly. Should the seller not agree to make the specific investment, they go to negotiation phase 2b.

5.1.2 Phase 2a: negotiation of supply price, for the case where the seller invests in the plant

In this phase, buyer and seller renegotiate face-to-face on price p for at most 10 min. If the renegotiation is successful and the two agree on the same or a new supply price, the experiment moves to phase 3a (market arrangement). Otherwise, with negotiation breakdown, the buyer earns 0 (zero) and the seller has a fallback payoff according to the specific investment treatment (explained below, in phase 3).

5.1.3 Phase 2b: negotiation of the assembly fee, for the case where the seller decides not to invest in the plant

In this case, the buyer herself makes the investment, either by acquiring the seller, or building the plant herself (I = 40). In the acquisition mode, the seller, now a hired employee, has to assemble the car for a fixed fee x. Buyer and seller are allotted 10 min to engage in face-to-face negotiations on x. Here however the assembly costs, which are a function of the time the seller takes to assemble the car, will be the responsibility of the buyer. This is done to implement a potential agency problem within the firm: when hired to assemble the car, the seller does not receive the corresponding profits from the operation, thereby potentially reducing his incentives to work efficiently. If they agree on the assembly fee x, then the experiment moves to phase 3b (vertical integration with hiring). If however there is no agreement on x, the seller earns 0 (zero), and the experiment moves to phase 3c, where, as a fallback option, the buyer herself can assemble the vehicle (vertical integration with own production).

5.1.4 Phase 3: Lego^® vehicle assembly

During this phase, either the seller or the buyer will assemble the car, depending on the outcome of negotiation phases 1 and 2. To assess production cost, which is a function of assembly time, we use a stopwatch. Accordingly, assembly cost is equal to an overhead of 5 points, plus 0.10 points for each second taken in assembly (i.e. total cost c = 5 + 0.10×s). The assembly should take no more than 10 min. In the event the vehicle is not assembled within that time, a session monitor (one of the authors, or an assistant) finishes the assembly and charges a penalty of 40 points. Buyers and sellers assemble the car separately, so buyers do not monitor the seller’s assembly activity.

5.1.5 Phase 3 involves three possibilities

In the market arrangement (3a), the seller invests in the plant and the seller assembles the vehicle. In this case, the buyer earns the sales price of the vehicle (v = 100 points) minus the supply price p paid to the seller. On the other hand, the seller earns the supply price p, minus the assembly cost c (a function of assembly time), minus the investment I (40 points). In the arrangement vertical integration with hiring (3b), the buyer acquires the seller, who assembles the Lego^®. Here, the buyer earns the price of the car (v = 100 pts), minus the assembly fee x paid to the seller, minus the plant acquisition I (40 pts), minus the assembly cost c (a function of assembly time). The seller in turn earns simply the assembly fee x. Finally, in the vertical integration with own production mode (3c), the buyer invests in the plant and assembles the car herself. In this arrangement, the buyer earns the price of the vehicle (v = 100 pts), minus the cost of assembly c (a function of assembly time), minus the plant investment I (40 points). The seller, on the other hand, earns 0 (zero).

5.2 Treatments

We implement a 2 × 2 × 2 factorial design with three treatments: distinct competencies, specific investments, and exchange conditions affecting perceptions of trustworthiness.

5.2.1 Distinct competencies

Using the Lego^® assembly task described before, we implement distinct competencies by allowing the seller (but not the buyer) to practice assembly at least twice for 15 min. In the treatment involving similar competencies, both buyers and sellers have an opportunity to practice assembly. We checked this manipulation by asking buyers to indicate at the end of the experiment the extent they agree with the following statement: “The seller was better trained than me to quickly assemble the car.” The treatment with distinct competencies had a significantly higher score (P < 0.001).

5.2.2 Specific investment

In line with the holdup mechanism discussed in the literature (e.g. Klein et al., 1978; Ellingsen and Johannesson, 2004), we manipulate specific investments by the extent to which the seller can recover his previous plant investment if the price negotiation breaks down during phase 2a. In the treatment with high investment specificity, the investment will be totally sunk (k = 1) and hence the seller’s fallback payoff when there is no price agreement is −40 points. In the treatment without specific investment, the plant is totally salvageable (k = 0), so if there is no agreement, the seller loses nothing. We checked this manipulation by asking sellers to indicate their agreement to the statement: “The seller, when investing in the plant, was at risk, given that he could have lost the invested value if he did not receive a good price for the assembled car,” with a higher agreement found in the treatment with specific investment (P = 0.095).

5.2.3 Trust

In experiment 1, we examine exchange conditions that influence perceptions of the partner’s trustworthiness. Our conditions affect distinct stages of the game, and follow established literatures. At a pre-investment stage, we manipulate whether buyers and sellers are allowed to meet and talk prior to the actual experimental decisions. Our dummy personal interaction equals 0 (zero) if sellers had to decide to invest or not in the plant without knowing the buyer’s identity ex ante, and 1 if sellers could freely interact and communicate with buyers before they decided to invest or not in the plant. In this treatment then, we allowed pre-investment personal interaction to unfold naturally, whereby parties could verbalize their concerns and non-binding promises. Previous theoretical and experimental work argued that pre-play communication promotes an alignment of goals and expectations regarding subsequent negotiation stages (see Farrell, 1988; Crawford, 1998; Ellingsen and Johannesson, 2004), thereby helping reduce (although not totally neutralize) the risk of opportunism. Note that in this treatment, trust is created before the seller makes the decision to invest in the plant, so it is expected to affect especially the perceived likelihood that the buyer will engage in holdup. In other words, as per H1, the trust in this treatment should have a negative interaction with specific investments and so attenuate the effect such investments have on the risk of vertical integration (see Ellingsen and Johannesson, 2004, for a similar empirical study involving holdup and pre-investment personal interaction).

We also ex post collected data on a different condition affecting perceptions of trustworthiness: the relational embeddedness of buyer–seller pairs (Granovetter, 1985; Ruef, 2002; Lazzarini et al., 2008). Although we do not formally manipulate this condition, we are able to assess the effect of heterogeneous expectations of trustworthiness by adopting Granovetter’s (1992: pp. 34–35) notion of relational embeddedness as an exchange condition that influences the parties’ perceptions of trustworthiness, insofar as “… how a worker and a supervisor interact is determined not only: by the meaning of these categories in a technical division of labor, but also by the kind of personal relations they have, as determined by a history of interactions.” Accordingly, we measure the relational embeddedness of each randomly assigned pair through a questionnaire administered right after the experimental session. We follow the “social connection” survey proposed by Glaeser et al., (2000: p. 820) in their trust experiment, in which both subjects answer together and try to reach a consensus to avoid measurement error due to relative perceptions of relational strength. (In the second experiment, we will also examine the effect of expectations of each party’s trustworthy behavior.) We ask a series of questions to capture the relationship history of the paired subjects as either strangers, acquaintances, or very close friends with recurring interactions. An exploratory factor analysis revealed three items that appear to effectively gauge embeddedness: whether they had met each other before; the extent to which they knew one another well (from “strangers” to “know each other very well”); and the strength of their friendship (from “strangers” to “close friends”). Cronbach’s α is 0.869, so we conclude they have a high degree of internal validity. We then averaged the items to create the score metric embeddedness.

Different from the previous case, this measure of trust is expected to affect especially the perceived likelihood that the seller will shirk in the assembly process after the decision to invest or not in the plant. After the initial investment stage, buyers and sellers meet to renegotiate price x, taking into consideration their preferences as well as perceived trustworthiness on the partner, as affected by the level relational embeddedness. Thus, as per H2, trust (embeddedness) should positively interact with the treatment distinct competencies, that is, distinct competencies will more positively affect vertical integration when buyer–seller pairs are in a high embeddedness condition.

5.3 Subjects

Our subjects are graduate (MBA) and undergraduate students from one public and two private universities. The majority is male (63.1%), with an average age of 26.7 years. In experiment 1, we paired up 350 subjects as buyers and sellers, for a total of 175 pairs. We randomly assigned subjects to each treatment cell, and to their roles as buyer or seller. Students were told to negotiate and maximize their own gains, but we gave no instructions as to whether they should deceive or play fairly. Because we have diverse students in our sessions, randomly assigned to distinct treatments, we are able to control for the potential effect on their previous experience, including previous work in organizations. Namely, we add in our regressions variables coding their age and course (MBA or undergraduate). This procedure increases our confidence that our results hold for a broad range of personal characteristics.

6. Results

We test our hypotheses using multinomial logit models, given the three possible organizational arrangements: the market arrangement (baseline option), vertical integration with hiring (i.e. alternate option, where the buyer acquires then hires the seller), and vertical integration with own production (i.e. the fallback option, where the buyer invests in the plant and assembles the car). In our regressions, we include not only the treatments but also a set of controls related to buyer and seller traits potentially affecting outcomes (age, gender, and graduate). These help us to control for remaining demographic differences across experimental conditions. We also estimate robust standard errors clustering by the location where each experimental session was run. Table 1 presents descriptive statistics and frequencies of organizational choices, while Figure 3 shows the percentage of choices according to distinct exchange conditions (the baseline choice is the market arrangement). Table 2, in turn, shows regression results.

Figure 3.

Experiment 1: summary of choices according to exchange conditions.

Open in new tabDownload slide

Though our central interest lies with moderating effects of specific investments and distinct competencies with trust, as per Figure 1, we start with analyses of main effects, in models 1 and 2 (Table 2). In these models, the coefficient of specific investment is not significant, while the estimate for distinct competencies indicates that this condition discourages vertical integration with own production (P < 0.05)—which is expected, given this is a fallback case where buyers would spend significantly more time assembling the car by themselves. Model 2 introduces the trust variables, showing opposite effects, both of which align with our hypotheses. While personal interaction largely reduces the likelihood of vertical integration (either with hiring or own production, P < 0.01), the effect of relational embeddedness is positive and promotes vertical integration with hiring (P < 0.10). These distinct effects can be explained by the fact that each of these trust variables mitigates distinct exchange hazards as explained before. The effect of personal interaction occurs at a pre-investment stage, so promotes higher investment in the specific plant, as sellers perceive lower holdup hazards in the market arrangement. In contrast, the effect of relational embeddedness occurs at a post-investment stage, after which parties freely negotiate the terms of trade and buyers can decide to hire their paired sellers or alternatively engage in own production (if they had not previously invested in the plant). Hiring the seller, as we argued, entails a risk of hierarchical failure instead, if the seller shirks in the production process; but trust (embeddedness) lowers this risk, creating an exchange condition that triggers vertical integration with hiring.

Yet formal testing comes by way of their interactions, that is, by examining how our trust variables moderate the effect of specific investment and distinct competencies. For instance, the non-significant effect of vertical integration in our previous regressions could be due to our previous theoretical prediction that its effect is highly contingent on trust, which is heterogeneous across our treatment cells. Following Hoetker’s (2007: p. 336) recommendation to examine interactions in non-linear discrete choice models, we implement split-sample regressions, through which we compare the main effects of our key trust variables across subsamples with and without each treatment. Accordingly, we ran models 3a, b, 4a, and b by splitting our sample in treatments involving the absence or presence of specific investments and distinct competencies, respectively, and examine the expected effects of the relevant trust variables operating in each case (i.e. pre-investment personal interaction in the case of specific investments and relational embeddedness in the case of distinct competencies).

As predicted, models 3a and b show a negative interaction between specific investment and trust (in the treatment personal interaction). Comparing the coefficients of personal interaction in subsamples 3a (non-specific investment) and 3b (specific investment) for the option of vertical integration with hiring we observe the effect is significantly more negative in the latter (P < 0.01 as per a Chi-square comparison of coefficients). This result is consistent with the visual pattern depicted in Figure 3 and also with our previous theory discussion, where we argued that trust should more likely reduce the propensity of vertical integration when investments are highly specific. For the fallback option of vertical integration with own production, personal interaction shows a negative effect only in the condition involving specific investment.¹ Thus, buyer and seller personal interaction seems to attenuate the perceived holdup hazards in the market arrangement, thereby reducing the risk of vertical integration (recall that in our setting interaction occurs at the very first stage where the seller is considering investing in the plant). All said, H1 is fully supported.

Also as predicted, models 4a and b show a positive interaction between distinct competencies and embeddedness (i.e. in our conceptualization, trust which is sparked in socially embedded relations). After the sellers’ initial decision to not invest in the plant, buyers and sellers are paired to renegotiate the hiring fee x. Thus, at this stage, the level of embeddedness is mostly important to help buyers anticipate whether sellers can be trusted to carry out the tasks if hired as an employee or whether they are likely to shirk, and so embeddedness helps determine whether buyers will hire sellers to assemble the car (vertical integration, hiring) or instead do it by themselves (vertical integration, own production). Also consistent with the pattern graphically shown in Figure 3, our estimates indicate that embeddedness indeed has a positive effect that is significantly higher in model 4b, involving a condition of distinct competencies, compared with model 4a, with similar competencies (P < 0.05). Therefore, trust, as indicated by the level of embeddedness in the relationship, enhance the effect of distinct competencies in fostering vertical integration with hiring, thus supporting H2.

7. Experiment 2

7.1 Experimental setting

Experiment 2 is designed to promote a more focused test of H2. In this new design, we abstract from market failure considerations, that is, we drop the first stage in experiment 1 where sellers consider whether to invest or not in a specific plant that could be potentially subject to holdup. Now, the buyer either outsources assembly for a given price or chooses to hire the seller. In line with our previous theoretical discussion, we now also introduce potential production synergies as the distinct competencies are vertically integrated. These synergies create an incentive to pursue vertical integration, but, as before, the gains from integrating distinct competencies depend on perceptions of trust. Thus, the final prediction is identical to our previous discussion leading to H2. (Due to space limitations, we leave the adjusted formal game-theoretical model corresponding to experiment 2 in the Supplementary Appendix SF) Each experiment session is broken down into three phases, described below (Figure 4).

Figure 4.

Experiment 2: sequential phases.

Open in new tabDownload slide

7.1.1 Phase 1: the buyer’s offer

In this phase, the buyer chooses between two options to get the seller to assemble the car: market arrangement or vertical integration with hiring. These two modes differ in important ways. In a market arrangement, the buyer offers a supply price p, paid to the seller in exchange for the assembled car. If the seller accepts this offer, he invests in the plant and pays for the assembly costs. Alternatively, in the vertical integration with hiring arrangement, the buyer pays for the investment, and offers a fee x to hire the seller to work as a regular employee in her firm. As in experiment 1, to ensure the buyer is not totally dependent on the seller, we also offer her a third (fallback) option, vertical integration with own production, whereby she can manufacture the vehicle herself.

7.1.2 Phase 2: the seller’s choice

In this phase, the seller decides whether to accept the buyer’s offer made in Phase 1. If he does, they move on to either phase 3a (market) or 3b (vertical integration with hiring). If he rejects the offer, they instead move to phase 3c (vertical integration with own production). To ensure the seller is not solely dependent on the buyer’s bids, we offer him the option to earn five points (plus his show-up fee) to exit the game at this phase. This option reflects an average profit sellers usually earned in our pre-experiment trials, and represents his opportunity costs.

7.1.3 Phase 3: assembly of the Lego^®

In this phase, assembly takes place, and the experiment ends. To compute payoffs and performance, we consider production costs as a function of variable (i.e. assembly time) and fixed costs (i.e. overhead). Variable costs involve 0.10 points per second, while fixed costs are 10 and 5 points, respectively in the market and hierarchy modes. We implemented this small difference to manipulate the synergies discussed in our formal model (as captured by parameter δ), and which usually reflects complementarities in complex, interdependent production processes (e.g. Kogut and Zander, 1992; Cremer et al., 2007; Zenger et al., 2011), Thus, total cost is c = 5 + 0.10×s under the vertical integration arrangement and c = 10 + 0.10× s under the market arrangement. Although internal production synergies should incentivize the choice for vertical integration (as we theoretically discussed earlier), our hypothesis center on how trust further promotes integration after these synergies are controlled for; if internal agency hazards are perceived to be high, parties may refrain from choosing vertical integration even when they could possibly exploit cost savings in an integrated production setting.

Note that in the arrangement involving vertical integration with hiring, the assembly costs are still a function of the time the seller takes to assemble; but in contrast to what happens in markets, they are now the buyer’s concern. As in experiment 1, this was done to implement a potential agency risk; that is, within the hierarchy the seller does not receive the corresponding profits from the operation, thus having reduced incentives to work efficiently. To further magnify the potential agency risk, we introduced an opportunity for the seller to engage in an unrelated task that he could voluntarily choose to complete. In this case, for five points, he could fill out an unrelated survey on his career goals. With this task, which takes on average 2 min and it is due before the car assembly is complete, the seller can essentially earn extra money at the expense of the buyer who in this hierarchy mode pays for the extra time.

7.2 Treatments

This new experiment involves a 2 × 3 factorial design with two treatments: distinct competencies and, as in experiment 1, and exchange conditions affecting perceptions of trustworthiness.

7.2.1 Distinct competencies

As in experiment 1, we implemented distinct competencies in a supply chain by considering the steep learning curve associated to the task, and allowing sellers (but not buyers) the opportunity to practice assembly of a Lego^® toy car for 15 min, prior to making decisions.

7.2.2 Trust

As in experiment 1, we manipulated two exchange conditions affecting a trustee’s perceptions of the partner’s trustworthiness. In a first condition, just like in the first experiment, we varied pre-play personal interaction, being that in the no interaction cell, we kept randomly paired buyers and sellers in different rooms, with no awareness of their partner’s identity, whereas in the interaction cell, partners were free to pair up and chat about their goals and intentions before the actual decisions. Notice that, different from experiment 1, the prediction in this case is that personal interaction will increase vertical integration, since it is expected to enhance buyers’ perceptions of the trustworthiness of the paired sellers.² In a second condition, while preserving anonymity, we allowed an additional form of interaction where sellers were asked to read and sign a code of conduct emphasizing expected trustworthy behavior. To guarantee understanding, experimental monitors read the code to all participants and asked sellers to hand write their names and sign the code in the same room, with all buyers still present. However, as in the baseline condition, buyers and sellers were still paired anonymously. After sellers signed the code, buyers were taken to a separate room, so that actual decisions and assembly tasks could commence. To avoid characterizing this manipulation as a determinant of contractual incentives, the code was non-binding; that is, there was no penalty if the seller signed the code and then cheated in their subsequent interactions. In the personal interaction treatment, sellers also signed the code of conduct.

In experiment 2, we also more directly assessed negotiators’ character trustworthiness (Gabarro, 1978; Colquitt et al., 2007). This variable is measured as the mean of six items on a 7-point Likert scale, and reflects concepts of honesty, transparency of intentions, justice, benevolence, concern, and dependability (Cronbach’s α is 0.880 so we are confident to have high internal validity). We built this measure by adapting scales from prior research in related settings (e.g. Cummings and Bromiley, 1996; McEvily and Tortoriello, 2011). The inclusion of a direct assessment of character trustworthiness (in addition to manipulating exchange conditions that induce trust to emerge between partners) makes it for a thorough assessment of how trust affects choices to vertically integrate. Because in this experiment sellers can act opportunistically when hired in the vertical integration mode, we surveyed the buyers’ perceptions of the seller’s character. Also, because another competing explanation is that such synergies may not be unrealized merely due to partners lacking technical skills for the job (Gabarro, 1978; Colquitt et al., 2007), we further add a control variable coding buyers’ confidence in the competence of their partner (which scholars often refer to as competence trust; Gabarro, 1978), measured as the average of three items, each on a 7-point Likert scale, regarding subjects’ expectations about the assembler’s abilities, dexterity, and aptitudes with Lego^® toys. Our metric of competence trust also shows high internal validity (Cronbach’s α = 0.884).

A battery of manipulation checks strongly confirms the validity of our design. In a post-experimental survey, we asked players about their level of agreement with the following statement: “In our pair, the person playing the role of seller was more capable of speedily assembling the toy than the one playing the buyer.” A mean comparison test reveals a significantly higher score in the distinct versus the similar competencies cells, for both buyers and sellers (P < 0.01). We also checked whether sellers who had an opportunity to practice assembly before the experiment in fact performed quicker assemblies than sellers in the experimental condition with similar competencies. A mean comparison test confirms the latter took significantly more time to assemble the cars (P < 0.01). Finally, mean comparison tests indicate that buyers’ scores for confidence in the competence of the partner are significantly higher in the treatment distinct competencies (P < 0.05), which again indicates that our manipulations were highly effective in creating competence differentials for sellers in that group.

As for the trust treatments, we also asked in this final survey if buyers agreed sellers had “fulfilled the spirit of the code of conduct.” Buyers in the code of conduct and personal interaction treatments showed notably higher agreement on this scale than those in the baseline condition where no code was applied (P < 0.01). More importantly, buyers perceived code-signatory sellers as displaying higher character trustworthiness (P < 0.01), while these sellers also less frequently cheated, that is, they more likely refrained from engaging in the unrelated task at the buyer’s expense (P < 0.05). We thus conclude that that our trust manipulations (both at the personal and group levels) were also highly effective.

7.3 Subjects

We recruited 456 subjects to experiment 2 (graduate and undergraduate students from the same universities), and randomly assigned them to the treatments as well as the roles of buyer or seller. We used 308 graduate (MBA) and undergraduate students (154 pairs) in main sessions, and other 148 (74 pairs) in extra sessions designed to assess assembly time (to be explained later). Our choice for these subjects follows recent research indicating this population to be a reliable source of data, similar to that of professional managers (Fréchette, 2015). In total, 58.8% of our subjects are male and, on average, 21 years of age.

8. Results

Table 3 brings descriptive statistics and correlations of key variables, while Figure 5 shows graphically the percentage of organizational choices according to experimental conditions. Part of the relative increase in the incidence of vertical integration, compared with experiment 1, can be explained by the presence of potential production synergies in the integrated mode (as explained earlier, we control for such synergies, and expect trust further affects the likelihood of vertical integration especially in the mode with hiring).

Figure 5.

Experiment 2: summary of choices according to exchange conditions.

Open in new tabDownload slide

Table 4 presents our econometric results. As in experiment 1, we model the determinants of vertical integration using a multinomial logit specification involving those three organizational options, where the market arrangement is again the baseline. Models 1 and 2 are baseline specifications with main treatment effects and trust variables, whereas models 3 and 4 test our hypothesis (i.e. how the effect of distinct competencies varies with trust) using split-sample regressions. We introduce character trustworthiness separately from the variables measuring exchange conditions affecting perceptions of trustworthiness (personal interaction and code of conduct) because, as explained previously, exchange conditions can affect the bases on which a person trusts another (e.g. in the no personal interaction treatment, pairs are not allowed to see each other and hence cannot form trustworthiness beliefs on each other). In other words, exchange conditions and character trust are positively correlated with each other.