2.1 Cheap talk

In the communication literature, pre-play “cheap talk” (i.e., non-binding, non-verifiable communication) is repeatedly shown to outperform many other mechanisms in resolving coordination or cooperation failures (e.g., Brandts & Cooper Reference Brandts and Cooper2007; Duffy & Feltovich Reference Duffy and Feltovich2002). In a large-scale meta-analysis of 130 distinct experiments from 37 studies, Sally (Reference Sally1995) finds that communication is the most effective mechanism for solving social dilemmas such as the public goods game and the prisoner's dilemma.

Economists are greatly interested in a theoretical understanding of the potential importance of cheap talk. Early studies in this area focus on whether pre-play communication can facilitate equilibrium play. For instance, Farrell (Reference Farrell1987) discusses how cheap talk can foster asymmetric coordination in an entry game and offers the following logic: if players’ announced plans constitute a Nash equilibrium, then this equilibrium would become focal and be followed by both players, thus reducing the frequency of ex-post disequilibria. Rabin (Reference Rabin1994) extends the analysis of Farrell (Reference Farrell1987) to unlimited rounds of communication. In his model, the players make repeated, simultaneous statements about their intended actions before playing a coordination game; if this communication between players is sufficiently long, each player should perform at least slightly better than her worst Pareto-efficient Nash equilibrium.

The analyses by Farrell (Reference Farrell1987) and Rabin (Reference Rabin1994) provide a useful benchmark for evaluating the effectiveness of cheap talk. Their arguments are supported by several follow-up experimental studies in which cheap talk is shown to increase both the incidence of equilibrium play and the frequency of efficient equilibria. For example, Cooper et al. (Reference Cooper, DeJong, Forsythe and Ross1989) investigate the effect of communication in the Battle of the Sexes game. They find that communication significantly increases the frequency of equilibrium play, as the players can indicate their planned actions using cheap talk messages. Duffy and Feltovich (Reference Duffy and Feltovich2002) find that one-way cheap talk results in increased coordination via a pure Nash equilibrium and thus increases efficiency. Blume and Ortmann (Reference Blume and Ortmann2007) study minimum and median games with Pareto-ranked equilibria in which multiple players can send numerical messages to signal their intended choices. Their results show that cheap talk increases coordination with a Pareto-efficient equilibrium.

However, no consensus exists regarding the effectiveness of pre-play communication for achieving efficient outcomes. Aumann (Reference Aumann1990) expresses doubt that cheap talk can enhance trust and promote efficiency in a stag hunt game, given that the intention to cooperate during a hunt is self-serving: each player strictly prefers to tell others to cooperate, regardless of her own intention. However, Charness (Reference Charness2000) shows experimentally that even self-serving cheap talk can facilitate coordination to achieve an efficient equilibrium, provided that the signaling stage precedes the action stage.

Increasingly, studies are exploring why cheap talk increases the efficiency of outcomes. Conventionally, scholars invoke behavioral heuristics to explain the effectiveness of communication. For example, Ellingsen and Johannesson (Reference Ellingsen and Johannesson2004) argue that people's concerns about fairness and consistency drive them to follow through on their commitments. Charness and Dufwenberg (Reference Charness and Dufwenberg2006) offer a similar perspective, positing that players are guilt-averse and thus wish to avoid failing to fulfil their commitments to others. Vanberg (Reference Vanberg2008) revisits the game of Charness and Dufwenberg (Reference Charness and Dufwenberg2006) and proposes that the result is driven not by guilt aversion but rather by an intrinsic preference for promise-keeping. Lying aversion, defined as a person's intrinsic motivation to maintain the credibility of their communications, is another common behavioral explanation for the effectiveness of cheap talk (Charness & Dufwenberg, Reference Charness and Dufwenberg2010; Ellingsen & Östling, Reference Ellingsen and Östling2010; Lundquist et al., Reference Lundquist, Ellingsen, Gribbe and Johannesson2009). Fehrler et al. (Reference Fehrler, Fischbacher and Schneider2020) find that while promises increase transfers between subjects, the credibility of a statement crucially depends on participants’ self-selection into cheap-talk situations.

2.2 Comparison of communication protocols

The economics literature compares various channels and structures of communication to determine the effectiveness of different communication protocols (see, e.g., Brandts et al., Reference Brandts, Cooper and Rott2019). Experimental evidence suggests that these protocols vary in terms of effectiveness. For example, Bochet et al. (Reference Bochet, Page and Putterman2006) find that efficiency in a public goods game increases markedly with communication via a chat room but does not increase with numerical communication via computer terminals. Similarly, in a collective resistance game, Cason and Mui (Reference Cason and Mui2015) observe a higher level of coordination when rich communication is used than when simultaneous intention signals are used. Ben-Ner et al. (Reference Ben-Ner, Putterman and Ren2011) compare two types of communication in a trust game: a single-round exchange of numerical proposals by a trustor and trustee, and a combination of proposal exchanges and unrestricted free chats. The authors find that although both kinds of communication facilitate trust and trustworthiness, verbal communication has a stronger effect than numerical signals. Charness and Dufwenberg (Reference Charness and Dufwenberg2006, Reference Charness and Dufwenberg2010) find that written free-form promises increase trust and trustworthiness significantly in a one-shot trust game with hidden actions. In contrast, the use of a pre-designed “bare promise” message (for example, “I promise to choose ‘Roll’ ”) only moderately enhances trustworthiness. Lundquist et al. (Reference Lundquist, Ellingsen, Gribbe and Johannesson2009) provide evidence that free-form messages lead to fewer lies than do pre-formulated promises in a hiring game with one-sided information asymmetry. In their experiment, the participants first perform a real-effort task to determine their skill, which is private information. They are then matched and play a one-shot game in which the applicant first signals her ability and the employer decides whether to offer a contract. The authors find that the proportion of low-skill applicants who over-report their abilities is significantly higher under the treatment with pre-formulated messages than under the treatment with free-form messages. Cooper and Kühn (Reference Cooper and Kai-Uwe2014) investigate how various communication types may foster cooperation in a two-stage collusion game. They find free-form chat to be the most effective communication method, especially when threats are involved. Moreover, they find that communication via a limited message space is not a good substitute for natural conversation.

2.3 What makes free-form communication more effective?

Ample evidence supports the superior efficacy of free-form communication compared with restricted messages, prompting researchers to investigate the underlying mechanisms of this efficacy. Experimental studies conducted across different environments offer explanations centering around distinct factors.

A group of existing research highlights the relationship-building function of free-form communication. A notable example comes from Mohlin and Johannesson (Reference Mohlin and Johannesson2008), where researchers attempted to explain the increased donation amounts in a standard dictator game when the recipient sends free-form messages. They designed a third-party communication treatment to isolate the content effect of free-form messages while eliminating the relationship-building effect. The authors find that the contributions from the relationship effects are comparable to the content effect in facilitating effective communication. In a study by Coffman and Niehaus (Reference Coffman and Niehaus2020), the focus is on a sales setting where sellers engage in free-form conversation to persuade buyers to increase their valuations of objects. They examine two pathways of persuasion: appealing to the buyer's self-interest and building rapport. The results show that sellers more frequently target buyers’ self-interest, and changes in self-interest explain a significant portion of persuasion. However, sellers are found to be most persuasive when they focus on rapport building.

The ability of free-form messages to facilitate type detection has also been extensively discussed in previous studies. In He et al. (Reference He, Offerman and Van De Ven2017), the authors explore the factors influencing the effectiveness of face-to-face communication in resolving social dilemmas. They tested several possible channels, and type detection is identified as playing a key role in increasing cooperation. The research also emphasizes the freedom to create promises in communication, as it not only enhances the commitment value but also aids in revealing the type. In a study by Ismayilov and Potters (Reference Ismayilov and Potters2016), the aim is to understand whether promise-making (internal consistency) or promise-receiving (social obligation) can explain the effect of promises on trustworthiness. The results of that study indicate that promises do not directly induce trustworthiness; rather, promises are important in signaling types, being more likely to be made by cooperators than non-cooperators. Turmunkh et al. (Reference Turmunkh, Van den Assem and Van Dolder2019) investigate the credibility of nonbinding pre-play statements about cooperative behavior using data from a high-stakes TV game show. To better understand the messages, they propose a two-by-two typology focusing on the conditionality and implicitness. The assumption of lying aversion is introduced, where potential defectors prefer statements that can be interpreted later as true. The authors’ analysis shows a link between statements with conditionality or implicitness and a reduced likelihood of cooperation, suggesting that malleability is a reliable criterion to differentiate types.

Additionally, there is a small but growing literature using the richness of free-form messages as an explanation for the effectiveness of free-form communication. Dugar and Shahriar (Reference Dugar and Shahriar2018) conduct stag hunt games to understand the efficiency-enhancing capacities of various types of messages. While intentional signaling messages fail to facilitate cooperation, the ineffectiveness of such messages can be ameliorated by adding reason-based components. Notably, the reason-based messages are found to be an effective efficiency-enhancing device similar to free-form messages. Therefore, Dugar and Shahriar (Reference Dugar and Shahriar2018) posit that the effectiveness of free-form communication can be explained by the sender's ability to specify the reason for choosing the cooperative behavior. Wang and Houser (Reference Wang and Houser2019) use a pure coordination game to examine differences in effectiveness between communication via intention signaling and communication via natural language. Free-form messages in this environment are found to include both signaled intentions and attitudes (i.e., the strength of a message sender's desire to have her message followed), and people respond to both intentions and attitudes when making decisions. The study finds that the difference in attitudes improves coordination significantly as the player indicating a weaker intention yields to the player indicating a stronger intention.

The psychology literature suggests that people are more inclined to engage in socially desirable behavior when their sense of self is evoked. This concept is rooted in the understanding that individuals often strive to maintain a positive self-image and are thus influenced by how their actions align with their personal identity. For example, Bryan et al. (Reference Bryan, Adams and Monin2013) provided participants with an opportunity to claim unearned money at the experimenters’ expense, using instructions that highlighted the identity implications of cheating (e.g., “Please don’t be a cheater”) versus focusing on the action (e.g., “Please don’t cheat”). The study found that participants in the “cheating” condition claimed more money than those in the “cheater” condition, a pattern evident in both face-to-face and online settings. This outcome suggests the power of language in influencing ethical behavior by appealing to individuals’ desire to maintain a positive self-image. Similarly, Bryan et al. (Reference Bryan, Walton, Rogers and Dweck2011) examined the impact of framing survey items to reflect a personal identity (e.g., “being a voter”) versus merely describing a behavior (e.g., “voting”). This study revealed that identity-based framing significantly increased voter registration and turnout, as evidenced by state records. These results highlight the role of self-concept management in motivating socially desirable actions. Chou (Reference Chou2015) investigated the effectiveness of various electronic signatures in comparison to handwritten signatures in deterring dishonesty. The study's findings across seven experiments consistently showed that electronic signatures are less effective in curbing dishonest behavior. This ineffectiveness is linked to their inability to evoke a sense of self-presence in the signer. Meta-analyses further established the reliability of these associations, emphasizing the role of self-presence in ethical decision-making. These studies collectively show that when the self identity is evoked, individuals are more likely to behave in ways that are congruent with their values and societal norms. This tendency to act in a manner that preserves a positive self-concept can significantly shape behavior, particularly in contexts where ethical or moral decisions are involved.

The focus of our work is closely related to the last two strands of literature. To understand the difference in effectiveness between free-form messages and restricted messages, we consider both the role played by the richness of natural language and the self-constructed nature of the messages.

3.1 Sequential social dilemma

We design a sequential social dilemma game involving two players: a seller and a buyer. In the game, the buyer takes initiative and offers a price, which can be either high ( $P_H$ ) or low ( $P_L$ ). Based on this offer, the seller then decides whether to provide a high-quality ( $Q_H$ ) or low-quality ( $Q_L$ ) product.

Both the buyer's evaluation of the product, denoted as V(Q), and the production cost, denoted as C(Q), increase with the quality of the product. Specifically, higher-quality products are of higher value to the buyer and impose higher production costs on the seller. To introduce a social dilemma element, we carefully selected the following set of parameters for our experiment: $V\left(Q_H\right)=300,V\left(Q_L\right)=100;P_H=200,P_L=50;C\left(Q_H\right)=100,C\left(Q_L\right)=0$ .

In the game, the buyer's payoff is calculated as the difference between the value of the product and the price offered, while the seller's payoff is derived from the difference between the price and the production cost.

Fig. 1

The sequential social dilemma game

Figure 1 presents the game tree, wherein a unique subgame perfect Nash equilibrium is achieved when the buyer proposes a low price and the seller, in response, provides a low-quality product. However, an opportunity for Pareto improvement exists if the buyer decides to offer a high price and, as a result, receives a high-quality product. This framework allows the examination of communication dynamics and their impact with respect to enhancing outcomes between the buyer and the seller across different treatments.

3.2 Communication treatments

Under the baseline treatment, the sequential social dilemma is conducted without any communication between the buyer and the seller. To compare the effects of free-form messages with those of restricted protocols, we introduce different communication treatments. Under these treatments, the seller is allowed to send a free, non-binding message to the buyer before the dilemma game begins. For this experiment, we designed four communication treatments that vary in terms of the form of the messages exchanged between the buyer and the seller.

Beginning with restricted messages, the restricted promise treatment allows the seller to choose between sending a restricted form of a promise, stating “If you pay a high price, I will choose high quality,” or leaving the message blank. Under the restricted threat treatment, the seller can choose to send a restricted form of a threat, stating “If you pay a low price, I will choose low quality,” or leaving the message blank.

Under the free-form communication treatment, the seller has complete freedom to write messages without any constraints on the number of words or message length.

Under the message-select treatment, the sender can choose from a list of 10 randomly selected messages crafted by sellers during the free-form communication treatment sessions, or leaving the message blank. Here, the richness of language in the message space (as analyzed in Sect. 5.4) is similar to that used under the free-form treatment, but the factor of message self-construction by the sender is removed. This allows us to isolate the effects of self-construction and message content richness that contribute to the differences in persuasiveness between restricted and free-form messages.

In all four communication treatments, we ensured that buyers were fully informed about the nature of the messages, whether they were selected from a pool or constructed by the sellers themselves. Under the two treatments with restricted messages, the buyer is fully informed about the messages available to the seller. Under the free-form treatment, the buyer is aware that the seller is free to write anything. Under the message-select treatment, the buyer is aware that the seller's messages are chosen from a set of previously constructed options. However, the buyer is not provided with the explicit set of messages available to the seller. For the message-select treatment, the available message options were limited to a pool collected from the free-form treatment. To balance variety with manageability, we provided a list of 10 randomly selected messages, aiming to avoid overwhelming subjects with an excessive number of choices. This approach, however, may have introduced potential bias, as subjects might not have found messages that precisely matched their intended communication. Nevertheless, this limitation is further discussed in the results section, where we compare the content delivered in free-form and message-select treatments, and their effectiveness.

By comparing the differences between treatments, we can also effectively study the prevalence of cooperation failure under the baseline treatment and replicate previous findings that free-form messages tend to better promote cooperation than do restricted messages. Furthermore, this design enables us to deepen our insights into the dynamics of free-form and restricted communications and their respective impacts on cooperation outcomes.

3.3 Experimental details

The experiment took place at the Smith Experimental Economics Research Center, Shanghai Jiao Tong University, in November and December 2019 and November 2020.Footnote ² The z-Tree program was used to conduct the experiment. Details regarding the numbers of sessions and participants per treatment can be found in Table 1. Overall, the experiment involved 354 participants, none of whom had prior experience with similar studies.

Table 1

Summary of the treatments

Treatment	Number of sessions	Number of subjects
Baseline	3	70
Restricted threat	3	70
Restricted promise	3	70
Message-select	3	72
Free-form	3	72
Total	15	354

We adopted a between-subject design wherein each subject experienced only one treatment. Before experiencing the treatment, the subjects were provided with printed instructions. To ensure better comprehension, a recording of the instructions was played aloud. The detailed instructions can be found in Appendix 3.

Under each treatment, the game was played repeatedly for 10 rounds. We used stranger matching to pair the subjects and assigned their roles randomly. Under the message-select treatment, the list of messages was re-selected randomly during each round and varied across the sessions.

To ensure a non-negative payoff, both the seller and the buyer were provided with 100 tokens at the beginning of each round. After each round, the players were informed of their counterpart's actions and earnings during that round. To prevent the players from hedging their bets across different rounds, payments were made only during one randomly selected round (for more details, see the discussion in Azrieli et al., Reference Azrieli, Chambers and Healy2018).

Upon the conclusion of the experiment, we collected additional information from the subjects, including their demographics and risk preferences, which were assessed using the method proposed by Crosetto and Filippin (Reference Crosetto and Filippin2013). In addition, we examined their social preferences as part of the data collection process.Footnote ³

The experimental sessions lasted approximately one hour on average. The subjects’ average accumulated earnings were 56.18 RMB. This amount includes a 15 RMB show-up fee. The U.S. dollar/Chinese yuan (RMB) exchange rate during the experiment ranged from approximately 6.54 to 7.07.

4.1 The seller's problem

We assume that the seller incorporates both material interest and psychological costs into his utility function:

$\begin{array}{c}\underset{S,Q}{max}{U}_S=P-C\left(Q\right)-\theta ·S·\psi (P,Q)-\omega ·\chi (S,Q)\end{array}$

where $\psi (P,Q)=\left(\begin{array}{cc}1& ifP=P_{H}andQ=Q_L\\ 0& otherwise\end{array}\right)$ , $\chi (S,Q)=\left(\begin{array}{cc}1& ifS>0andQ=Q_L\\ 0& otherwise\end{array}\right)$

There are two psychological forces at play in this context. The first force, guilt aversion, is represented quantitatively as $\theta ·S·\psi (P,Q)$ . Here, $\psi (P,Q)$ is a binary variable that takes a value of 1 if the sender fails to meet the buyer's expectations, and 0 otherwise. In the game, this occurs when the buyer offers a high price but receives a low-quality product, leading the seller to experience a disutility level of $\theta ·S$ . The parameter $\theta$ reflects the seller's sensitivity toward guilt and can take one of two values: ${\theta }_H$ or ${\theta }_L$ , where ${\theta }_H$ is greater than ${\theta }_L$ . Both players share a common prior, denoted as $p\left({\theta }_H\right)$ . The parameter S denotes $\mathrm{sincerity}$ , which reflects the perceived persuasiveness of a statement.

The second force, aversion to lying, is denoted as $\omega ·\chi (S,Q)$ . Here, $\chi (S,Q)$ is a binary variable that takes a value of 1 if and only if the seller signals positive sincerity but ultimately supplies a low-quality product, irrespective of the buyer's choice. In such cases, a lying cost of $\omega$ is incurred.

4.2 The buyer's problem

After receiving the seller's message and detecting $\mathrm{sincerity}$ , the buyer forms her belief about the seller's type as $\mu \left(\theta \right|S)$ (referred to as $r_S$ hereafter) and then chooses a price that maximizes her expected utility:

$\begin{array}{c}\underset{P}{max}{U}_B=\sum _{\theta }\mu \left(\theta \right|S)·\left(V\right(Q_{\theta }(S,P))-P)\\ \end{array}$

$Q_{\theta }(S,P)$ represents the seller's optimal quality decision. For the analysis, we adopt the pure-strategy perfect Bayesian equilibrium as the solution concept.

We introduce two additional assumptions. First, following the approach of Ellingsen and Östling (Reference Ellingsen and Östling2010), we assume that the buyer possesses a weak lexicographic preference for being kind, such that $P_H\succ P_L$ when the buyer is indifferent. This assumption implies that we can always specify the buyer's exact choice. Second, we assume that the difference between the guilt sensitivity of the two types is significantly large, such that ${\theta }_H>C\left(Q_H\right)-C\left(Q_L\right)-\omega >{\theta }_L>0$ .

4.3 Equilibrium analysis

In this subsection, we discuss the equilibrium analysis for various communication protocols. These protocols differ primarily in the $\mathrm{sincerity}$ S of the messages that they permit. A detailed proof of the equilibrium analysis can be seen in Appendix 2.

Restricted Promise: Consider $S\in \{0,1\}$ , where $S=1$ if the seller chooses to send a restricted promise and $S=0$ if the seller sends an empty talk. In this case, the optimal strategy profile on quality is:

$\begin{array}{c}{Q}_{\theta }(S,P)=\left(\begin{array}{cc}{Q}_H& ifP=P_H,S=1,\theta ={\theta }_H\\ Q_L& otherwise\end{array}\right)\end{array}$

There are two pooling equilibria in this scenario. In the first equilibrium, both types pool on $S=0$ , and the buyer's best response is $P_L$ , regardless of the received message, when $r_0\in [0,\frac{3}{4})$ . Consequently, both types of sellers respond according to $Q_{\theta }(S,P)$ and play $Q_L$ . This is the empty-talk equilibrium.

In the second equilibrium, both types of sellers pool on $S=1$ , and the buyer always chooses to trust,Footnote ⁴ selecting $P_H$ if the prior $p\left({\theta }_H\right)\in (\frac{3}{4},1]$ . Consequently, cooperation $(P_H,Q_H)$ can be achieved between a seller of type ${\theta }_H$ and the buyer, whereas a seller of type ${\theta }_L$ chooses to betray the buyer's trust. This is the limited sincerity pooling equilibrium.

Message-select: Let us consider $S\in \{0,1,...,N\}$ where N is not sufficiently large, such that $\frac{C\left(Q_H\right)-C\left(Q_L\right)-\omega }{{\theta }_L}>N\ge 2$ . In this case, the seller must choose between using empty talk or sending messages with limited but distinguishable levels of sincerity ( $S>0$ ). Here, the seller's optimal quality decision is

$\begin{array}{c}{Q}_{\theta }(S,P)=\left(\begin{array}{cc}{Q}_H& ifP=P_H,S>0,\theta ={\theta }_H\\ Q_L& otherwise\end{array}\right)\end{array}$

In this scenario, the equilibria are qualitatively similar to restricted promise communication. We observe an empty talk equilibrium and multiple limited sincerity pooling equilibrium where $S_{{\theta }_H}=S_{{\theta }_L}=\alpha$ for all values of $\alpha$ in the range $N\ge \alpha >0$ . Similarly, cooperation $(P_H,Q_H)$ can only be achieved between a seller of type ${\theta }_H$ and the buyer in those limited sincerity pooling equilibria.

Free-form Communication: Consider $S\in \{0,1,...,N\}$ , where N is sufficiently large: $\exists$ $N\ge k>2$ such that

$\begin{array}{c}1+\frac{C\left(Q_H\right)-C\left(Q_L\right)-\omega }{{\theta }_L}>k>\frac{C\left(Q_H\right)-C\left(Q_L\right)-\omega }{{\theta }_L}\end{array}$

There exist some statements that are sufficiently sincere ( $S\ge k$ ) that even a seller of type ${\theta }_L$ keeps their promise. Here, the seller's optimal quality decision is

$\begin{array}{c}{Q}_{{\theta }_H}(S,P)=\left(\begin{array}{cc}{Q}_H& ifP=P_H,S>0\\ Q_L& otherwise\end{array}\right)Q_{{\theta }_L}(S,P)=\left(\begin{array}{cc}{Q}_H& ifP=P_H,S\ge k\\ Q_L& otherwise\end{array}\right)\end{array}$

In this case, in addition to the multiple limited sincerity pooling equilibria wherein $\forall S$ in $k-1>S>0$ , there are two other types of equilibria where both types of sellers cooperate with the buyer.

The first type is the high sincerity pooling equilibrium, where both types of sellers tend to pool on highly sincere statements. Here, $S_{{\theta }_H}=S_{{\theta }_L}=\alpha$ for all values of $\alpha$ in the range $N\ge \alpha >k$ . The buyer's best response is $P_H$ , given $N\ge S\ge k$ , and $P_L$ otherwise. Both types of sellers ultimately achieve $(P_H,Q_H)$ .

The second type is the high sincerity separating equilibrium, where the two types of sellers send different highly sincere statements. In this equilibrium, $N\ge S_{{\theta }_H}=\alpha \ge k$ , $N\ge S_{{\theta }_L}=\beta \ge k$ , where $\alpha$ and $\beta$ are distinct. Similarly, the buyer's best response is $P_H$ , given $N\ge S\ge k$ , and $P_L$ otherwise. Both types of sellers can achieve $(P_H,Q_H)$ in this equilibrium.

Restricted Threat: In our framework, sincerity is linked to the psychological costs associated with guilt aversion and aversion to lying. This suggests that messages intended to promote cooperative behavior should signal positive intentions or show a certain level of commitment. However, the restricted threat that a sender can employ, which underscores the non-cooperative outcome ( $P_L,Q_L$ ), lacks such a commitment. Consequently, it's reasonable to assume that $S=0$ in scenarios involving this kind of threat. Therefore, in treatments with a restricted threat, we expect that both sellers and buyers will exhibit behavior similar to what is observed in the baseline scenario.

4.4 Hypotheses