Effects of Conversation Politeness on Hiring Decision in Online Labor Markets: An Inverted U-Shaped Relationship Exploration

Abstract

This study examined the effect of politeness, as a key reflection of linguistic features of conversation in the online labor marketplace, on hiring behavior. Drawing on the politeness theory, a non-linear relationship was theorized. A hypothesis was put forward and examined against a large-scale archival dataset from a Chinese online labor market. Using an econometric model, the results demonstrated that there was an inverted U-shaped relationship between politeness and hiring decisions. The study offers theoretical implications to the online labor market literature and politeness theory by providing empirical insights on the role of politeness in hiring decision. In addition, our findings offer beneficial and practical contributions for vendors and platform operators.

1. Introduction

An increasing number of online marketplaces, such as Taobao and Freelance, embed live chat functions into their platforms to strengthen consumer relationships and improve service experiences on these platforms. By simply clicking a “chat” button, consumers can initiate one-to-one synchronized communication with merchants, which reduces information asymmetry [1]. A recent industrial report conducted by Forrester Research suggests that approximately 42% of US consumers think that real-time communication with merchants make it easier for them to buy products or services that have not been experienced [2]. Additionally, it is also considered beneficial to merchants. For example, Sun et al. [3] demonstrated that if a seller replies to all consumer requests through online communication, their conversion rate can be increased by 9%. With regard to purchase decision, Tan et al. [1] showed that sellers can increase the purchase probability of tablets by 15.99% if they use live chat in their services. Given the crucial role live chat plays in online marketplaces, it has attracted wide attention from both scholars and practitioners.

Our comprehensive review indicates that the extant literature primarily focuses on the impacts of live chat on consumer behavior and business performance [1,4], while limited attention has been paid to the significant effect of live chat content itself. As Hong et al. [5] pointed out, the exploration of effects aroused by textual content is a rising but limited explorative field of IS research. Towards this direction, in this study, we sought to examine whether an especially important linguistic feature of conversation, i.e., politeness, could affect hiring decisions in online labor markets. We pursued this because, in online labor markets, the conversation is not only a way to communicate with the purpose of price negotiation, but it is also used to help employers to build a sense of whether a vendor they are communicating with is suitable to cooperate with [1,4]. In previous studies, Hong et al. [5] suggested that a polite vendor can allay an employer’s concerns about possible unpleasant experiences when cooperating with them, thus winning the favor of potential employers. Accordingly, exploring how politeness in conversation influences hiring decisions is crucial, and our study is expected to extend the current knowledge in this area.

This study makes the following contributions to the current literature. First, as shown in Appendix A, the emerging literature on the online labor market has mainly focused on exploring the role of the reputation mechanism in mitigating information asymmetry [6,7,8]; it has scantly studied the linguistic features of live chat. This paper is one of the very few initial attempts to explore these in the focal context of online labor markets. Second, by investigating the inverted U-shaped relationship between politeness and hiring decisions, this study contributes to the politeness theory [9,10]. Third, previous studies on live chat have ignored the role of linguistic features in online communication processes. This research gap is emphasized by prior works such as Hong et al. [5], in which the authors implied that the exploration of linguistic features of textual contexts is a rising area of IS research. To date, our work constitutes one of the first attempts to explore the influence of these linguistic features.

The remainder of this study is structured as follows: first, we review prior work and put forward a hypothesis. Next, we report the research context, variable measures, methodology, and model specification. Then, we present the estimation results and employ a multiple method to further check the robustness of the results. Last, we discuss the findings and implications, and present the limitations and future works.

2. Theory and Hypothesis

With the popularization of communication technology, an increasing number of online marketplaces have integrated live chat functions into their platforms to strengthen consumer relationships and facilitate transactions on these platforms [3]. Live chat refers to a web-based function that provides a channel through which consumers can communicate synchronously with merchants to reduce information asymmetry [1]. Recently, several studies have started to explore the role live chat plays in online marketplaces. For example, the relationships between the usage of live chat and consumer behavior [4,11], attitudes [12,13], conversion rates [3], and purchase probability [1] have been widely studied.

These studies have reported diverse and increasingly rich findings. Most of them have emphasized the impact of live chat on traditional e-commerce, but few have studied this in the context of online labor markets. Online labor markets, such as Freelance and Upwork, are digital platforms that offer employers and vendors the ability to enter into contracts for intangible service products, such as software design, brand logo design, computer programs, etc. [5,14]. Service products are highly customized and project-specific, and are produced after the contract is awarded (i.e., service-related products take shape after purchase) [5,6,15]. These characteristics lead to information asymmetry on both sides, that is, the vendor does not know the demands of the employer, while the employer does not know whether the vendor can fulfill their demands. Considering the highly complex nature of service products, it is understandable that social factors are likely to play a very important role in these relationships. This is highlighted by Hong and Shao [16], who suggested that the completion of service products is inseparable from ongoing communication between employers and vendors. In addition, live chat tools are mainly used for understanding product specifications and price negotiation in traditional e-commerce platforms [1,4]. However, these tools may serve a great yet subtle role in online labor markets, as the completion of a service product relies on the cooperation of both parties, and thus employers may try to evaluate from the communication process whether a vendor will be easy to cooperate with. This theory has been confirmed by previous studies, such as by Hong et al. [5], which suggested that linguistic features are a prerequisite for cooperative work.

As the key reflection of linguistic features of dialogue, politeness is the central concern of previous researchers. Sociolinguistics often regard politeness as a social norm or a set of normative social rules [17]. It is a central force in communication that promotes social interaction and largely determines how effective the communication is. The politeness theory offers an appropriate lens to interpret such a phenomenon [10,18]. The theory suggests how individuals choose appropriate words to create and maintain the feeling of closeness to achieve communication goals. In recent years, politeness theory has attracted wide attention from IS researchers. For example, Lee et al. [10] studied the implication of politeness theory in the context of knowledge management platforms. They argue that polite answers are more likely to be accepted by the asker and vice versa. Hong et al. [5] suggest that the benefits of direct messaging for employees depend largely on the employees’ politeness. Liu et al. [19] suggest that politeness can convey positive emotions. Accordingly, politeness is considered to be a key factor of business in online labor markets, as successful production of a service product is inseparable from the continuous communication and cooperation between employers and vendors [5]. That is, a polite vendor may allay an employer’s concern about unpleasant experiences in project cooperation and win the favor of potential employers in the process of communication. Conversely, if a vendor cannot treat his/her potential employers politely, they will lose them gradually.

According to the politeness theory, politeness may influence the sense of closeness only within an acceptable range of politeness [18,20]. That is, social norms stipulate the acceptable range of politeness in a specific situation, within which a speaker may choose to be much politer and thereby increase the sense of interpersonal closeness. Once beyond the range, politeness may create a sense of social distance. Previous studies found the negative effects of excessive use of linguistic cues. For example, when employers encounter content with excessive politeness, they may realize that the vendors are trying to convince them by exaggerating or even lying [21]. This is also highlighted by works such as Zhang et al. [22], where excessive linguistic cues lead to untruthfulness and even convey the feeling of deception. Accordingly, excessive politeness will make employers feel unreality, which makes it difficult to smoothly communicate with the vendors and increase the sense of social distance. Thus, it is conjectured that politeness in vendor’s chat content could generate an inverted U-shaped relationship with employers’ hiring decision. Therefore, we propose the following hypothesis.

Hypothesis. 

Politeness has an inverted U-shaped effect on employers’ hiring decision.

3. Research Methodology

3.1. Research Context and Data Collection

We collected data from Zhubajie, a leading online labor market in China, which attracted over 31 million registered members by the end of 2021. The business model of this platform offers to connect potential employers and vendors for service-product transactions, such as logo design, brand marketing, software development, and so on [23,24]. The data were collected in January 2019 and pertained to 7952 orders.

As shown in Figure 1, the platform has its live chat component, which is functionally similar to instant-messaging applications, such as Skype. This component is embedded in the website, and its icon is displayed on the service homepage. When potential employers click the icon, it will be automatically activated and immediately brought them into the chat dialogue box. In addition, the platform has recorded of detail chat conversations, such as the message content, timestamps of each message, and the information about users who use live chat, which provides a good foundation for our research. For each conversation, if an employer signs a contract with the vendor, the chat records would be selected before the pre-contract conversation; otherwise, all chat records are selected. Our analysis focused on pre-contract conversation because post-contract conversation, as the name suggests, could not influence the employer’s hiring decision and was beyond the scope of our research.

3.2. Measures

Hiring decision. The dependent variable in our study was whether an employer makes a hiring decision (henceforth HD) after the conversation, which was operationalized as a binary variable. In line with prior research, such as Hong et al. [5], we coded the dependent variable HD as 1 if the employer made a hiring decision with the vendor, and as 0 if otherwise.

Politeness Density. The key challenge in assessing the influence of politeness on hiring decision is how to quantify the vendor’s politeness. In line with extant literature, we used several methods, such as text mining, human labeling, and machine learning, to construct a measure for politeness [5,10]. Specifically, a vendor’s messages were split into words with a python package Jieba (a Chinese word segmentation tool) [25]. Given that the politeness of language is inherent to function words rather than content words, it is possible to judge the extent of politeness by constructing an established polite dictionary [26]. In line with Lee et al. [10], we firstly constructed linguistic markers based on the Chinese Dictionary that were likely associated with politeness. Thereafter, we counted the occurrences of the politeness markers and divided the count of the politeness markers by the number of words or phrases to obtain the density of politeness markers in pre-contract conversation.

$$politeness=\frac{{\displaystyle \sum _{i=1}^{i}politenes{s}_i}}{n}$$

(1)

Control Variables. This paper controlled a set of factors that might influence employers’ hiring decision, such as authentication, sales amount, review rating, tenure, year, prior experience, and amount of information. First, authentication refers to the confirmation of vendor’s real address information on the platform. In this platform, vendors are not forced to authenticate the real address. In line with Zhang et al. [27], it was operationalized by whether the vendor has passed real address authentication (a dummy variable). Second, sales amount could be directly derived from the database by calculating the cumulative amount of transactions of the vendor prior to the focal transaction [28]. Third, review rating (the average rating of a vendor on the platform on a 5-point Likert scale) was included in our analysis because it has been found to affect consumer decision [16]. In line with Lin et al. [29], we employed the average review rating of the vendor prior to the focal transaction. Fourth, tenure was measured by the vendor’s registration duration. Fifth, year was measured by the vendor’s registration duration. Sixth, prior experience has been treated by prior studies on social relationships as a sign of actual trust between vendor and employer [30]. It is both a knowledge-based antecedent of current trust and a proxy of actual trust in these prior transactions. According to Hong and Shao [16], we operationalized it in terms of whether an employer had transactions with the vendor on this platform prior to the focal transaction (a dummy variable). Seventh, Lv et al. [4] suggest that a larger amount of information can increase the employer’s detailed understanding of the services. Thus, we operationalized it by the total number of replies that a vendor offered to the employer prior to the focal transaction. Control variables 1–4 were included because they were all vendor attributes displayed on the live chat box, which could be easily observed by employers to influence their hiring decision. Control variables 5–7 were included to capture the employer’s characteristics or perception, which were considered to be important variables affecting decision-making in previous studies, such as Lv et al. [4] and Hong and Shao [16].

We defined our key measures and provided a definition for each variable in

Table 1. Description of key variables.

Variable	Abbreviation	Description
Dependent variable
Hiring decision	HD	Hiring decision as 1 if the employer makes a deal with the vendor, and as 0 if otherwise.
Focal variables
Politeness density	PD	Politeness density is calculated by dividing the count of the politeness markers by the number of words or phrases.
Control variable
Authentication	AU	Dummy variable: 1 if the vendor is verified and 0 otherwise.
Sales amount	SA	The amount of sales the vendor has prior to the focal transaction.
Review rating	RR	The average numerical rating of reviews the vendor has received from the employers on completed works (5-point Likert scale).
Tenure	Tenure	The number of years the vendor has been registered on the platform.
Year	Year	The number of years the employer has been registered on the platform.
Prior experience	PE	Dummy variable: 1 if the vendor and the employer have had transactions on this platform prior to the focal transaction and 0 otherwise.
Amount of information	AOI	The total number of replies that a vendor offers the employer.

.

3.3. Data Analysis and Results

We used STATA 15 to perform our estimation.

Table 2. Descriptive statistics of study variables.

Variable	Mean	Std. Dev.	Min	Max
HD	0.139	0.346	0	1
PD	0.052	0.064	0	1
AU	0.121	0.326	0	1
SA	1,360,478.3	3,485,227.3	0	20,411,970
RR	4.826	0.797	0	5
Tenure	3.154	4.764	0.071	49.033
Year	1.877	2.023	0.003	12.06
PE	0.022	0.146	0	1
AOI	14.412	17.817	1	292

and

Table 3. Correlation matrix.

Variables	VIF	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
(1) HD		1.00
(2) PD	1.02	0.11	1.00
(3) AU	1.11	0.05	−0.07	1.00
(4) SA	1.07	0.07	0.07	0.18	1.00
(5) RR	1.15	0.05	0.07	−0.20	0.09	1.00
(6) Tenure	1.11	0.12	0.01	0.18	0.01	−0.29	1.00
(7) Year	1.02	−0.01	0.04	0.06	0.11	−0.01	0.01	1.00
(8) PE	1.01	0.25	0.02	0.04	0.05	0.02	−0.01	0.06	1.00
(9) AOI	1.03	0.29	−0.03	−0.07	−0.06	0.10	−0.04	−0.04	0.08	1.00
Mean VIF	1.07

displayed the descriptive statistics and correlation matrix, respectively. As our variables of politeness density, authentication, sales amount, review rating, tenure, year, prior experience, and amount of information were power-law distributed, we employed a natural logarithm transformation [3,31]. The correlation coefficients among these variables were less than 0.7; the VIF values for all variables were less than 3 (range from 1.01 to 1.15), with an average value of 1.07, thus, multicollinearity did not seem to be an issue in this paper [32,33].

Because our dependent variable was a binary variable, according to prior research, such as Lv et al. [4] and Coussement et al. [34], we used the binary logistics regression model to analyze the data collected from ZBJ. For better interpretability, all variables were scaled to z-standardized values ($z(x)=(x-\overline{x})/sd(x)$). The regression model was as follows:

$$\begin{array}{l}logit[P(H{D}_j=1|X_i)]={\beta }_0+{\beta }_1\cdot P{D}_i+{\beta }_2\cdot PD^2_i+{\beta }_3\cdot A{U}_i+{\beta }_4\cdot S{A}_i+\\ {\beta }_5\cdot R{R}_i+{\beta }_6\cdot Tenur{e}_i+{\beta }_7\cdot Yea{r}_i+{\beta }_8\cdot P{E}_i++{\beta }_9\cdot AO{I}_i+{\epsilon }_i\end{array}$$

where $i$ represents the orders, $P(H{D}_j=1|X_i)\in \left\{1,0\right\}$, 1 represents that the employer makes a hiring decision, and 0 if otherwise; control variables include authentication, amount of sales, RR, tenure, year, prior experience, and amount of information.

The results were presented in

Table 4. Regression results.

	Dependent Variable: Hiring Decision
	M1	M2	M3
PD		6.952 ***	8.269 ***
		(0.589)	(0.848)
PD squared			−9.036 *
			(4.577)
AU	0.299 **	0.426 ***	0.436 ***
	(0.107)	(0.110)	(0.110)
AS	0.0316 ***	0.0282 ***	0.0273 ***
	(0.00529)	(0.00536)	(0.00537)
RR	1.159 ***	1.095 ***	1.082 ***
	(0.244)	(0.248)	(0.248)
Tenure	0.889 ***	0.862 ***	0.854 ***
	(0.0687)	(0.0696)	(0.0698)
Year	−0.105 +	−0.131 *	−0.129 *
	(0.0609)	(0.0618)	(0.0618)
PE	2.829 ***	2.835 ***	2.839 ***
	(0.182)	(0.188)	(0.189)
AOI	1.090 ***	1.159 ***	1.146 ***
	(0.0470)	(0.0489)	(0.0494)
_cons	−6.288 ***	−6.454 ***	−6.380 ***
	(0.447)	(0.459)	(0.461)
Observations	7952	7952	7952
Wald chi2	840.60	896.70	887.96
Log pseudolikelihood	−2666.6941	−2606.544	−2603.9493
AIC	5349.388	5231.088	5227.899
BIC	5405.238	5293.919	5297.71
Pseudo R2	0.1675	0.1863	0.1871

Robust standard errors in parentheses. *** p < 0.001,** p < 0.01, * p < 0.05, ⁺ p < 0.1.

. Model 1 merely incorporated the control variables; Model 2 analyzed the impact of politeness on hiring decision in conjunction with control variables. Model 3 added the squared term of politeness. In addition, the significance of regression coefficients of all variables did not change, indicating that the results were robustness. As shown in Model 3 of Table 4, the results showed that politeness ($\beta =-9.036,p<0.05$) has an inverted-U relationship with sales, which supported our hypothesis.

4. Discussion

In online labor markets, live chat tools alleviate information asymmetry between employers and vendors in the transaction process and improve operation efficiency. This is especially important for service products whose product quality could not be well predicted before delivery [7]. The politeness theory helps to build the whole theorization. Based on a large scale online chatting and transaction data, as predicted, we found that politeness has an inverted U-shaped relationship with hiring decision. The findings of this study have important implications for the research as follows.

First, research regarding live chatting tools in online labor markets is in its infancy. As shown in Appendix A, the extant literature on online labor markets has mainly focused on exploring the role of reputation mechanism on mitigating information asymmetry [6,14,35]. Although these studies have yielded rich research results, the role of live chat content has been significantly ignored. Considering the customized and project-specific nature of service products, its transaction is inseparable from the constant communication between employers and vendors [5,6,15], live chat is thus expected to play an irreplaceable role. Up till now, the present study is one of the very initial attempts, thus extending the current knowledge.

Second, this study contributes to the politeness theory. Politeness theory, from a psychological perspective, is used to interpret how individuals choose appropriate linguistic words to create or maintain the feeling of closeness in order to achieve their communication goals (Brown and Gilman 1989; Lee et al., 2019). However, scholars point out that excessive linguistic cues will lead to counterproductive effects [21,22]. These inconclusive findings indicate that the relationship between politeness and the likelihood of communication goals may not be linear. Our research investigates the relationship between politeness and hiring decision. The results show that the relationship between politeness and hiring decision is a non-linear inverted U-shaped relationship, thus extending the current knowledge of the focal theory.

Third, this study extends the emerging literature on live chat usage in online markets. The emerging literature on live chat mainly focuses on two streams, with: (1) the direct effect of using live chat [1,3]; and (2) objective aspects of live chat content, such as message length and message counts [4,36]. However, the role of linguistic features in live chat process has been scantly studied. This is also highlighted by prior research, such as Hong et al. [5], where the author suggests that the exploration of linguistic features of textual context is an exciting area for IS researchers. Integrating politeness theory, this paper generates new insights into the role of linguistic characteristics and responds to the call for more research into the role of live chat in the online marketplace [3].

Meanwhile, this paper also yields several crucial insights for both vendors and platform managers. In online marketplaces, live chat tools are often used to negotiate prices and understand a product [1,4]. However, beyond that, it may play a great and subtle role in online labor markets, since the completion of service products requires the cooperation of the employer and the vendor [5]. Our results reveal an inverted U-shaped relationship between politeness and hiring decision. Thus, when vendors contact potential employers, they should ensure that they use a polite tone as the employers may try to infer whether the vendor is easily working with after signing the contract from the communication process. In addition, vendors should also be cautious that politeness has a double-edged sword effect. Excessive politeness may make employers feel unreality, which makes them difficult to communicate smoothly with the vendors and increase the sense of social distance.

Moreover, this study provides a number of actionable insights for operators of online labor markets. For platform operators, it is necessary to further improve the performance of the live chat function to ensure that employers can contact vendors in time when consulting, in order to promote subsequent interactions and online transactions. Specifically, the platform operators may consider the following aspects: first, the current design of the live chat is a passive trigger mode; that is, the vendor can interact with the employers only if a request is initiated by the employer [1]. Considering the importance of the live chat in promoting service transactions, a proactive live chat may be a useful way for vendors to identify potential employers [37]. Second, the platform operators should develop a real-time semantic analysis tool or language reminder function to enable vendors to grasp the degree of politeness in time. Third, platform operators might coach vendors on effective linguistic communication strategies through the provision of guidelines.

Limitations and Future Research

This study is also subject to several limitations, which offer opportunities for future research. This paper primarily centers around the archived data collected from the online labor markets, which may, to some extent, hinder the generalizability of findings. As suggested, the online labor market is essentially differing from traditional e-commerce platforms, such as Amazon, in many crucial ways [5,23,38]. Therefore, it is unclear whether our research conclusions can be applied to generalizable platforms, such as Amazon. Broader context-oriented studies could be conducted to explore rich findings.

In addition, despite the present study has employed a hierarchical regression model to confirm the effectiveness of the result, some explanatory variables beyond our control may affect employers’ hiring decision. Conceivably, an employer may communicate with multiple vendors during the actual service product transaction process. Due to the limitations of the data in this paper, it is difficult to control the impact of this potential competition on employer’s hiring decision. Therefore, we call for future research to consider the competition among vendors, so as to extend the current knowledge and provide more practical guidance for platform operators.