Master Thesis 2013-2014 University of Amsterdam
Grammar errors, spelling errors and typos in e-mail communication: do customers care?
Wouter Werensteijn
Student number 10640444
Master Business Studies – Marketing track
1
Index
Abstract ... 2 Introduction ... 2 Literature Review ... 6 Complaint satisfaction ... 6 Loyalty ... 7 Word of Mouth ... 9 Service Quality ... 9 Service Recovery ... 10 Communication Quality ... 14The Five P’s Standards Model and The Customer Perspective... 15
The Conceptual Model and Hypothesis ... 20
Methodology ... 21 Sampling ... 21 Design... 22 Procedures ... 24 Measures ... 25 Results ... 27 Analysis steps ... 28 Discussion ... 35 Conclusions ... 36 Limitations ... 38 Future research ... 39 List of References ... 41 Appendices ... 46 Appendix A ... 46 Appendix B ... 47 Appendix C ... 51 Appendix D ... 53 Appendix E ... 54
2
Abstract
Mistakes in written communication are often made and may dissatisfy customers. Especially when a company is trying to resolve an error made. This research examines the effect of the total of grammar errors, spelling errors and typos (GSaT) on complaint satisfaction in e-mail communication. Using an experimental scenario design based on the five P’s model, rating e-mail communication, Dutch participants between 18 and 65 familiar with e-e-mail
communication were asked to answer a survey to gather quantitative data. The effect of GSaT on Complaint Satisfaction was tested using one hypothesis and no other variables. The results of this research indicate that GSaT have a significant effect on complaint satisfaction
depending on the amount of written errors. This shows that to explain complaint satisfaction in e-mail communication, GSaT is a variable that cannot be underestimated. This knowledge can be used to enhance customer service in business.
Introduction
Companies often lack in the complaint management area. Customers often receive no response after their e-mail complaints and when they do receive an answer, it is often incomplete or not answering questions asked. After speaking to people, it became clear that health insurance companies receive complaints from customers for not being transparent about health care costs in the invoices to customers. Hospitals and other health care providers in the Netherlands have three years to invoice health care costs. By the time some of these providers invoice, a customer does not know why they have to pay the invoiced amount of money. In the same complaint to the company, the customer also complains about not knowing what compensation can be expected for several necessary future treatments. The insurer answers the latter complaint, but does not explain anything about transparency of the
3
costs made. This may result in an unsatisfied customer that still does not have an answer to all the questions. How often does it occur that customers do not get proper responses from companies when something went wrong and why does this happen? These questions have many possible answers, as communication is a complex process in which many things often can, and in fact do go wrong.
According to Appelbaum (1973), communication is a critical interpersonal process. It consists of a repetitive cycle of initiating, maintaining and terminating information exchange. Also, communication is not a unidirectional process. The scope of the effect of a
communicative action is broad because there is not only an effect on the receiver, but also on the sender after the receiver responds to the sender. This means that the role of the receiver is
more active than one would think(Schoop et al., 2010). For example, a customer complains
about a product that does not work properly because he or she wants a working product. The company responds positively and sends a new product to the customer. The company shows that it wants the customer to get what was paid for and wants to leave a positive impression. The customer feels satisfied with the service and the company can start using a working product. Thus, looking at the explanation of the word ‘communication’ and the accompanying process, the complexity becomes clear.
Because we are living in a time in which computer technology is dominating and developing fast, it is increasingly valuable to thoroughly investigate digital complaint
management communication media. When considering the expansion of commercial internet use, it is logical that complaints are sent and handled through e-mail communication
(Dickinger & Bauernfeind, 2009). Especially because the internet is becoming increasingly relevant for the product or service delivered. Some examples of digital media for companies are: e-mail, chat session with a web care employee via the company website, message posting via Twitter and message posting via the Facebook page. However, not all companies use these
4
communication forms yet. Companies, like people, change over time. Also, the way companies use technology changes, but change takes time. The format that almost every company uses is e-mail communication. E-mail communication still is the number one way for companies and consumers to communicate with each other. E-mail has made it easier for customers to make contact with companies. E-mails sent by customers to companies, mainly consist of enquiries and complaints (Matzler et al., 2005). Perhaps this is because of the crystal clear advantages of this type of written communication over other types of written and spoken communication. Most importantly, there is more time to answer a complaint for the sender, information is stored digitally so it does not need to be remembered and information is easily accessible for the receiver. In addition, e-mail communication has a ‘speed’ advantage for both parties and message delivery costs are very low compared to other available media. Therefore, this research focuses on e-mail communication to customers and particularly on how customers evaluate the total of grammar errors, spelling errors and typos (GSaT). In previous studies, there has not been done any testing on this topic.
5
For this research, academic insights are acquired about the customer perspective regarding e-mail communication and the resulting (dis)satisfaction after e-mail
communication with companies. The focus is on GSaT to see what the impact is of this ‘written’ factor in e-mail communication on complaint satisfaction. Figure 1 shows the research.
Figure 1. The research
To address the research gap, which will be further explained at the end of the
literature review, the following main research question will be answered: What is the impact of the GSaT quality measure, in e-mail communication, on complaint satisfaction?The accompanying academic relevance will also be explained at the end of the literature review. However, the value in practice can be made clear already. Managers are able to make positive changes in training courses for employees to make sure that employees’ written
Complaint through
e-mail Company
(complaint management)
Answer via e-mail using known influences on complaint satisfaction and GSaT Customer Complaint satisfaction Impact?
6
communication performance improves significantly. To conclude, managers will also know if customers care about GSaT, so they are able to check how the company is performing on this quality measure.
The outline of this research is as follows: first an extensive literature review, conceptual model and the associated hypothesis will be presented in which the important factors in this research will come forward. This will be followed by the methodology used to conduct the research. Results will follow from the data that were gathered. Finally, the research concludes with a discussion about the conclusion, the limitations of the research and future research possibilities.
Literature Review
To evaluate existing research of impact of written communication quality (in specific GSaT) in e-mail communication on complaint satisfaction, theory is presented. The provided theory is also used to help fill the research gap and show the academic relevance. First, the customer satisfaction domain is unravelled. After this, the service recovery domain will be explored. Then, the complaint management model that brings everything together is explained. The literature review ends with the conceptual model and the matching hypotheses.
Complaint satisfaction
A distinction needs to be made between two different customer satisfaction constructs, namely satisfaction after a complaint (complaint satisfaction) and overall satisfaction with a company (cumulative satisfaction). Complaint satisfaction is the degree to which the customer perceives the company’s complaint handling performance to meet or exceed the expectations.
7
Overall satisfaction is the accumulated experience with the general performance of a company (Gelbrich & Roschk, 2011; Wendel et al., 2011). Allthough these constructs of satisfaction are related, they are distinct because some of the factors influencing them may differ (Shankar et al., 2003). Oliver (1997) clarifies the satisfaction after complaint handling by explaining that this form of satisfaction can be referred to as transaction-specific satisfaction. This is the judgement of a single observation or transaction. In this research transaction-specific
satisfaction is key because transactions can be seen as critical incidents. Pointing out these incidents through complaints, customers can make companies aware that something went wrong and might keep going wrong.
It can be intuitively understood that this transaction specific satisfaction has a clear impact on the cumulative satisfaction. However, from the meta-analysis study by Gelbrich and Roschk (2011), it follows that complaint satisfaction has no significant impact on the cumulative satisfaction when estimating the path model. Looking at the seven studies however, five of them agreed that the impact is significant as is the correlation between the two forms of satisfaction in the bivariate analysis. According to the authors, justice
perceptions, which are the individual subjective assessments of organizational responses, predominate in explaining the cumulative satisfaction when competing with transaction-specific satisfaction in an overall path model. However, this statement does not mean that transaction-specific satisfaction is not essential for companies.
Loyalty
Bodet (2008), explains that the two forms of satisfaction seem to be more
complementary than competitive, albeit that cumulative satisfaction seems to be a better predictor of loyalty. Still, this means that a link from transaction-specific satisfaction to loyalty does exist. The relationship between satisfaction and loyalty however, is asymmetric.
8
Satisfied customers do not automatically become loyal customers (Oliver, 1999). Shankar et al. (2003), explain that loyalty consists of two constructs, namely attitudinal and behavioural loyalty. Behavioural loyalty may be spurious because customers stay with a company until there is a better alternative in the marketplace. Attitudinal loyalty means that customers have some attachment or commitment to a company and are not easily swayed by a more attractive offer. Attitudinal loyalty indicates higher repurchase intent, resistance to counter-persuasion, resistance to adverse expert opinion, willingness to pay a price premium and willingness to recommend the company to others. Considering both constructs, attitudinal loyalty contains most value because there is some attachment or commitment to a company and this type of loyalty indicates different benefits for companies. Therefore, the focus is limited on attitudinal loyalty in this research. As said before, the relationship between satisfaction and loyalty is asymmetric. However, the possibility exists that high satisfaction levels increase loyalty and along with that important consequences such as repurchases and the likelihood of
recommending the company to others or in other words, positive word of mouth
communication. In the study by Gupta and Zeithaml (2006), other relationships come forward as well, so the authors acknowledge aforementioned consequences. The overall image of their study is that satisfaction increases the financial performance of companies, which is
understandable after previous explanation. Also, increasingly competitive markets make it more important to preserve loyalty and develop long-term relationships with customers (Blodgett et al., 1997). Loyal customers are therefore of vital importance to companies. Without loyal customers, companies may lose their right of existence because of insufficient revenues.
9 Word of Mouth
Focussing on word of mouth communication, when customers are not satisfied, there is the possibility of negative word of mouth. This means that customers tell others about their unsatisfactory experience. Word of mouth communication is perhaps the most important consequence of loyalty according to Black and Kelley (2009). Their study suggests that word of mouth is the most trusted source of information among customers. The more dissatisfied customers become, the more likely they are to negatively express their displeasure using word of mouth communication. Studies report that unsatisfied customers will tell between eight and ten people about bad service. One in five unsatisfied customers will tell twenty people.
However, with the increased popularity of social media, it can be expected for this effect to be even bigger today. The harm generated for companies through social media can be massive (Hocutt et al., 2006). There are many examples worldwide of customers who put their complaint on forums, the Facebook page of the company and/or on Twitter, etcetera. One such forums is the America OnLine’s ‘Wine and Dine Forum’. This forum regularly attracts complaints about restaurants (Hocutt et al., 2006).
Service Quality
In the study by Parasuraman et al. (1988), it becomes clear that communication is one of ten overlapping dimensions that consumers use to assess service quality. In the study by Sharma and Patterson (1999), it comes forward that communication is an important ingredient to achieve high perceived service quality. It is also stated that communication effectiveness is considered to impact on the two fundamental components of service quality, namely functional and technical quality. Functional quality is about how service is delivered and perceived by the customer, while technical quality is about what service is delivered and how it is perceived by the customer (Kang, 2006; Sharma & Patterson, 1999; Wendel et al., 2011).
10
In terms of Zeithaml (1988), service quality is the comparison customers make between their expectations and their perception of the service they received. Straus and Hill (2001)
acknowledge this and mention that e-mail communication is of great importance because proper interaction with customers, increases the customer’s service quality assessment. Therefore it is important to take e-mail communication serious. When service quality is not what customers expect it to be, complaints are likely to occur. According to Goodwin and Ross (1990), it is of great importance how companies respond to customer complaints. The response will affect consumers’ choice to do business again with the same company in the future. Wendel et al. (2011) agrees by saying that evidence from studies shows that quality of service is an important determinant of customer satisfaction. When customer satisfaction is low, future loyalty is unlikely and a potential future customer may be lost. Companies should therefore make sure that the service quality is superior. It is essential to achieve the highest customer satisfaction possible, especially after service failure(s). A service failure that leads to a complaint can be seen as a critical incident. It can change the relationship between two entities, in this case the relationship between a customer and a company. So, delivering quality service is considered an essential strategy for success and survival in today's competitive environment (Zeithaml et al., 1996), especially when it is a service recovery attempt.
Service Recovery
Service recovery refers to how a company responds to service failure. It occurs when a company attempts to make up for service failure. However, attempting to recover is not the same as recovering with a successful outcome. When the recovery is effective, customers’ expectations have been met or exceeded (Black & Kelley, 2009). Looking at different studies about service recovery, there are different opinions about the existence of the ‘service
11
recovery paradox’. It is a subject with contradicting findings (McCollough et al., 2000). The service recovery paradox means that customers rate a service higher/feel stronger
commitment to a company after a failure and the associated service recovery attempt than if no failure would have occurred (Black & Kelley, 2009; Hocutt et al., 2006). Most studies before the year 2000 and the study by Hocutt et al. (2006), conjecture that service recovery can increase satisfaction significantly above a failure-free service. An example is the study by Hart et al. (1990). It shows that when loyalty is not that high, effective handling of complaints can diminish the negative consequences of the company’s failures and chances are that more loyalty is created during the recovery than if the original transaction went well. The study by Singh and Wilkes (1996), acknowledges this by saying that well handled complaints can enhance loyalty and even customer retention. This is in contrast with more recent studies (McCollough et al., 2000; Michel et al., 2008). These studies show that a failure-free service, even when the service recovery is excellent, in general leads to the highest satisfaction possible. This means that the paradox does not exist or only exists under some conditions. McCollough et al. (2000); Michel et al. (2008) only find evidence for the paradox to exist, when the initial service is just ‘satisfying’ and the service recovery attempt is ‘much better than expected’. This is the only outcome that supports the presence of a paradox in relation to satisfaction.
As the latter studies indicate, it is of importance how superior the recovery is and of what quality the failure-free service is. Furthermore, the kind of failure and the harm failure can do to the customer plays an important role too. McCollough et al. (2000) use the example of an improperly prepared steak in a restaurant. When the recovery attempt is ‘much better than expected’, the steak is replaced, the customer does not pay for the steak and the waiter is responsive and empathetic. The harm done to the customer would be relatively low and the recovery attempt would completely mitigate any harm done because of the failure. But what if
12
the customer had to wait for another steak that required such an amount of time that the customer is late for a next appointment? Looking at all aforementioned about service recovery, it seems that nothing beats a high quality failure-free service, except a superior service recovery when the failure-free service is not ‘much better than expected’. However, the type of failure and the harm that can be done with it must not be underestimated.
Companies that deliver service attempt to reduce failures. However, failures occur because they are inevitable. For these service companies, recovery is of utmost importance. A service has a unique nature and therefore it is impossible to guarantee a failure-free service. Of course, companies should strive to identify and eliminate all potential sources of failure before the service is provided. This will maximize companies’ financial performance and customer satisfaction (McCollough et al., 2000). But when this complex process of improving service is not executed in an effective way, the employee can never be very successful.
Furthermore, customers will be disappointed every time and this enhances the possibility of spreading negative word of mouth and switching to competitors. This will eventually lower the increase in total future customers and lead to higher costs to acquire new customers (Goodwin & Ross, 1990; Gruber et al., 2009; Tax et al., 1998). When considering that it costs five times as much to attract a new customer than it does to retain a current customer
(Zeithaml, 2000), the need for companies to provide flawless service becomes even clearer. Customer retention is critical to profitability. In certain circumstances, a company can make profits go up to a hundred percent by increasing customer retention with a mere five percent (McCollough, et al., 2000).
An atmosphere that encourages dissatisfied consumers to seek redress therefore is preferred (Zeithaml, 2000). According to McCollough et al. (2000), service recovery can help to gain access to superior market intelligence about customer dissatisfaction so that
13
to identifying problems and actions to ensure that such failures do not happen again. Thus, seeing a complaint as a gift instead of a burden, can improve the service of companies significantly when action is undertaken accordingly. When the service becomes superior, the customer and the company will both benefit. Johnston (2001) emphasizes this by telling that previous studies have shown that excellent handling of complaints can significantly influence customer satisfaction. However, in the study by Hocutt et al. (2006), it comes forward that customers who do not complain about a service failure, are the biggest problem for
companies. It also comes forward that extensive research shows few customers complain about service failures. Customers do not file a complaint because they perceive companies to not be responsive to complaints, have a certain attitude toward complaining; are not
sophisticated enough to complain and find that not everything (level of product matters) is complaint worthy. For the likelihood of complaining, it also matters if the problem is stable or has been controllable and that customers do not know how to complain. However, according to Hart et al. (1990), many customers that complain, still end up feeling more negative about a company because of the way their problems are (not) handled. Thus it makes sense that people do not complain that easily.
Complaint management is the area within service recovery where complaints are handled. Complaint management is about the use of strategies by companies to resolve and learn from service failures in order to restore company’s reliability in the eyes of the
customers (Hart et al., 1990). In the words of Johnston (2001), complaint management is the process that handles complaints and recovers customers. The study by Harrison-Walker (2001) proves the shortcoming of complaint management with a study about the content that was posted on a internet complaint forum. Reasons for customer complaints were mainly all about the employees of a company. According to the complaints on the forum, employees were often rude, incompetent and gave misinformation. It shows that communication often
14
plays a big part in complaint management and thus it seems that effective communication is very difficult to achieve. Even when, according to Krapels and Davis (2003), many employers specifically identify communication skills as a job requirement. There is never a guarantee that effective communication will be used by those employees that possess the
communication skills the employer asked for. Communication skills are therefore of great importance for job applicants and career success, but no guarantee for effective
communication on the job and success in handling complaints.
Communication Quality
So what can be considered to be effective communication, or in other words, high
communication quality? To start, no meaningful definition of communication quality can be based on a one-dimensional interpretation of quality. Different authors have presented
definitions of communication quality for various research areas. Schoop et al. (2010) brought them together. Of the twelve definitions, five are useful for this research. This will be
explained later on when the model will be explained. The five definitions are:
• ‘Communication quality is communication that is positive, intimate, and in Control’; • ‘Communication quality is timely, accurate, adequate, complete communication’; • ‘Communication quality is communication effectiveness’;
• ‘Communication quality is the ability of the seller to communicate in seller-buyer relationships according to the buyer’s expectations’
• ‘The assessments of the quality of communication are a function of the completeness, credibility, accuracy, timeliness, and adequacy of communication flows’.
Looking at these definitions, they have to do with ‘how’ a response is built up. However, how the response is send, is of great importance as well. In the studies by Murphy and Gomes (2003); Murphy and Tan (2003); Strauss and Hill (2001), it comes forward that customers
15
prefer a fast response from a company, even if it is just a confirmation e-mail that the
complaint has been received and will be taken care of soon. So to reach high communication quality, written communication, in particular e-mail communication, is preferred. Raciti and Dagger (2010) acknowledge this by saying that in services characterised by ongoing
interactions, written communication is likely to be an important aspect to develop and maintain relationships with customers. Also, in the study by Strauss and Hill (2001), seventy business students respond with an average score of 5,6, on a seven point Likert scale, that e-mail is a good way for companies to communicate with customers.
The Five P’s Standards Model and The Customer Perspective
Studies in which e-mails are sent to companies to check how they perform in e-mail
communication on different quality measures, show that companies are often not responding to enquiries and complaints in a customer satisfying way. A study on Singaporean travel agencies (Murphy and Tan, 2003) shows that only one in four agencies responds to e-mails. From these, about one in ten use all eight quality measures suggested by the authors. The eight quality measures are: (1) send a reply within 24 hours, (2) open the e-mail with dear, (3) thank the customer for e-mailing, (4) politely close the e-mail, (5) address the customer by name, (6) include the employee’s name in the e-mail, (7) answer the question in the e-mail
and (8) include the company’s name in the e-mail. The same authors came up with the five
P’s standards model on how to respond to e-mail communication. This model was inferred from extensive research on online communication in the Swiss hotel industry by Murphy et al. (2003). Four of the five P’s were already used in this study. Only the P for ‘Promotional’ had to be added to complete the five P’s model. The eight quality measures in the
Singaporean travel agency study are grouped under one of the five P’s. The first P stands for ‘Prompt’ which consists of quality measure (1). The second P stands for ‘Polite’ which
16
consists of quality measures (2), (3) and (4). The third P stands for ‘Personal’ which consists of quality measures (5) and (6). The fourth P stands for ‘Professional’ which consists of quality measure (7). Translated to this research it means: respond to content of the complaint. The last and fifth P stands for ‘Promotional’ which consists of quality measure (8).
In contrast to the Singaporean travel agencies study, the study about the Austrian hotel industry (Matzler et al., 2005), shows improved response rates and response times of business in e-mail communication to customers. However, it must be stated that this study has been conducted later, in a different industry and a different culture. These factors surely explain the difference in response rates and response time since e-mail has become more important for companies.
Another important study is the ‘follow-up’ study on educational institutions in Australia (Murphy & Gomes, 2003). An updated five P’s standards model from Murphy and Tan (2003) is utilized in this study. This model is of great importance, which is acknowledged by Strauss and Hill (2001). Their study indicates that companies that choose to use e-mail for customer communication, can improve customer satisfaction by focusing on response
timeliness, personalization and other customer relationship development factors. The study shows that among the participants, quick responses to complaint e-mails enhances customer satisfaction. The quick response is achieved using a confirmation e-mail, to buy time for a more thorough response. Looking at personalization, e-mail responses should specifically mention the problem of the customer to show that the company is concerned about the
customer and the situation that led to the complaint. Next to this, e-mail responses should use the name of the customer. Also, the employee should use its own name when signing for the company. It is more personal, it gives the customer a name for future communication and the company can audit more easily for performance measures. It becomes clear that the focus in this study is on the P for ‘Prompt’ and the P for ‘Personal’. The study does not clearly say
17
anything about the other three P’s. Still, the study does check for quality measures that were later used by Murphy and Gomes (2003); Murphy and Tan (2003) to check for politeness, professionalism and promotional elements in e-mail responses. The importance of the P’s existence is clear, but the impact of the separate P’s is unknown, because a regression analysis has never been done in previous studies.
Looking further into the updated model, it consists of twelve quality measures divided among the five P’s. This means that four quality measures are added to the model. The P for ‘Prompt’ is still the same: (1) a reply within 24 hours. The P for ‘Polite’ is still the same: (2) open the mail with dear, (3) thank the customer for mailing and (4) politely close the e-mail. The P for ‘Personal’ is renewed: (5) address the customer by name and (6) give the
customer a personalised answer instead of a standardised answer. The P for ‘Professional’ is
renewed: (7) include the employee’s name (measure switched from ‘Personal’ to
‘Professional’), (8) include the company’s name (measure switched from ‘Promotional’ to ‘Professional’), (9) answer the question such that no follow-up e-mail is necessary (adapted measure), (10) avoid attachments in the e-mail (new measure). The P for ‘Promotional’ is renewed: (11) use a branded e-mail address for e-mail contact and (12) include a signature
file in the e-mail. The authors also mention that grammar is employed under the P for
‘Professional’. However, they do not specify why and without testing it in their study on educational institutions in Australia. In this study, the authors use a fictitious Chinese student and his father. They both send e-mails separately to educational institutions in Australia with the same content and questions, to see if the companies respond and how when adhering to different quality measures. Again the outcomes do not show high quality service
communication as it should be according to the quality measures from the five P’s standards model. However, all the companies are responding to the e-mails this time.
18
So why do Murphy and Gomes (2003) talk about grammar being part of the P for ‘Professional’? And why do they not use it as a quality measure? The latter question remains unclear since there may be different reasons. It would be likely that grammar is employed under the P for ‘Professional’. Grammar has everything to do with syntax (all rules used to specify the structure of sentences) to provide semantics (the meaning of sentences) in
language and thus with delivering a message. In addition, syntax and semantics can be used to assess communication quality (Schoop et al., 2010). Since the core of the P for ‘Professional’ is answering to the content of the complaint, grammar is very important. The study by Black and Kelley (2009) mentions that well written stories, using proper grammar and punctuation, give credibility to the storyteller, whereas poorly written stories make readers think that the writer cannot be trusted. In the study by Mirchandani (2012), it can be read that Indians are characterized as uniformly having poor grammar. Therefore, training is needed because it is important for service agents to deliver appropriate service. Fausey and Matlock (2011) find that the grammatical wording of political messages affect voting behaviour. This includes judgements whether candidates will be elected. Results show that imperfective (was VERB + ing) descriptions in comparison to perfective (VERB + ed) descriptions, have significant more impact. Bohannon and Stanowicz (1988) find that adults ignore children's speech errors. Adults focus more on well-formed sentences than language mistakes. However, adults also repeat with changes, or request clarification of a sentence containing syntactic or phonological errors. The results indicate that all adults tend to respond differentially to children's mistakes. Looking at these four studies it becomes obvious that grammar often matters and studies about it over the years, have been scarce. Studies about grammar in e-mail communication could not be found at all. The same goes for studies about spelling errors and typos in e-mail communication. Thus, it is important to check if and how much, customers care about the quality measure GSaT in e-mail communication. So the research gap can be considered to be
19
the missing of knowledge about the quality measure GSaT in e-mail communication. Also, it is of great importance to think about how to improve this ‘written’ factor. Describing a proper e-mail response is easy, but implementation is hard (Murphy & Gomes, 2003; Murphy & Tan, 2003).
Most studies, including aforementioned studies about e-mail communication, focus on a company perspective while it is very likely that differences exist between what customers and companies value important. Spenner and Freeman (2012) find that consumers want different things from companies through online media, in comparison to what companies expect from consumers. Consumers want discount and information while companies think that consumers interact with them online because they want to feel connected to the brand through a community. Thus consumers aim for transactions only while companies want to build a relationship. Translating this information to quality measures in e-mail
communication, it becomes highly possible that the quality measures companies consider important in e-mail communication, differ from the customers view of important quality measures in e-mail communication. Having said this, the academic relevance becomes clear. Customers and companies do not always think alike and so complaint satisfaction may not reach high(er) levels. Confirmation of this is shown in the work by Strauss and Hill (2001). The study however, is an exception because it is the only study that addresses the customer perspective to check for satisfaction levels of e-mail communication in handling complaints. It shows how students from American universities rate the e-mail communication quality for companies, addressing their complaints. When looking from the customer perspective at the different quality measures, company concern for customers and company credibility, it shows that companies deliver less service compared to what customers expect. Customers always expect an answer but almost half of them do not get a response. Customers also expect
20
companies to care about them and be credible however, companies lack in caring for their customers and are not good partners when it comes to credibility.
The Conceptual Model and Hypothesis
Based on the provided foundation of information, the conceptual model can be presented:
Figure 2. The Conceptual Model.
In Figure 2. ‘The Conceptual Model’, the use of the five P’s Standards Model can be seen. Also the hypothesis that will give insight in the GSaT and complaint satisfaction relationship, is made visible.
Independent variable
GSaT (total mistakes)
Dependent variable
Complaint satisfaction H1
Quality measure
Prompt
Functional service quality Technical service quality
Quality measure Polite Quality measure Personal Quality measure Promotional Quality measure Professional
21
H1: the more grammar errors, spelling errors and typos, the lower complaint satisfaction. Customers may think that companies do not care about them, or take them serious when written language in e-mail communication is not of high quality. To make sure this can be tested successfully, it is primarily important that the significance of GSaT is proven. Therefore the five P’s standards model will be used as presented. When the five P’s are being used as suggested by previous studies, GSaT will be the only quality measure that matters and may be able to ensure greater complaint satisfaction. How H1 is tested, can be read in the chapters ‘Methodology’ and ‘Results’.
Methodology
Now that necessary theory has been presented, the practical side of this research becomes point of interest. Therefore a detailed description about sampling, the design, the procedures and the measures used for this research, is given.
Sampling
The population consists of 10.679.287 Dutch people between 18 and 65 years of age as that were living in The Netherlands as of January 1, 2014 (CBS StatLine, 2014). However, not everybody from the population was suited for this research because familiarity with e-mail communication was a requirement. Today, still not everybody uses e-mail, especially not everyone from fifty years and older. From this population, a non-probability (convenience and snowball) sample was used. Participants came from multiple networks that consist of a broad range of people looking at age/gender/education. Sample size was almost impossible to measure since next to around one hundred initial invitees, people were invited by participants
22
as well. Based on communication from initial invitees, to let me know if they spread the surveys, the sample size was probably around 500 people. Still, the response rate was
unknown. However, an estimate would be a response rate around 45 percent. This is based on the familiar relationships with most of the participants that were initially invited for the survey and the relationships that exist between participants and the people they invited. In total 223 participants were almost evenly divided over the surveys and gathered in two weeks time. Although it is impossible to say what number of participants make up for a decent dataset, the more participants, the higher the reliability of the outcomes would be. In this research, the total of participants that was acquired seems to be high enough to assure increased precision in estimates of various properties of the population. The results should therefore be acceptable with the right usage of statistical methods.
Design
The research was done using a cross-sectional 4x1 between-subjects scenario-based survey design. Participants were randomly assigned to one of the survey versions, thus the scenario-based survey design can be considered an experimental design. In total, there were four survey versions because previous studies do not explain which amount of errors is superior in finding the tipping point where customers’ satisfaction levels drop. So, to check for the impact that GSaT has on complaint satisfaction, multiple levels of the independent variable GSaT needed to be tested. So, the first three versions contained an e-mail in which the independent variable (GSaT) varied. Survey [1] contained twenty mistakes, which can be thought of as a large amount in one message one could say. However, this was needed to make sure that GSaT has an impact as quality measure. Survey [2] contained ten errors. This still seems not very realistic, but it represents the middle between twenty and zero mistakes
23
and it has half as many mistakes than Survey [1]. Survey [3] contained five errors, half as many mistakes than Survey [2], which can be seen as realistic scenario. Especially looking at the daily practice and the overall not highly educated employees that work in customer service departments. The errors used in the first three surveys were a good balance between striking and more difficult to see errors, which are surely not easily found by the average person. Survey [4] contained an e-mail that was written mistake-free. This means that Survey [1], [2] and [3] were treatment groups while survey [4] was the control group. To be clear, all surveys were linked to H1, to check for difference in complaint satisfaction levels.
From Appendix A (Appendix B in Murphy & Tan, 2003), it can be seen what a high quality company response through e-mail looks like. However, this e-mail misses the four quality measures from the updated five P’s standards model. Therefore, the e-mails used within the surveys, were moderated through extending the example e-mail with the four quality measures from the model. Note that in the e-mails, an even more personal approach is used because of the ‘telephone component’ that is offered at the end of the e-mail. Thus effects that were found, existed because of GSaT. All other factors were accounted for by using the work of aforementioned authors. The different e-mails can be found in Appendix B.
It is important that attitudes about GSaT in e-mail communication were measured between four groups of participants at one point in time. This way, the time to take the survey was short and participants were not biased because they could not compare the different e-mails. Otherwise, a carry-over effect could have significantly influenced the results. However, a larger sample was needed because of the between-subjects design and the people within groups could have differed so that results would be hard to compare. So, in this research the scenarios were of great importance because they allowed control over otherwise
24
otherwise would have unfolded over a longer time period. The key disadvantages of participants not understanding the scenarios and participants not having the ability to understand a complaint situation, were solved through clearly and in short outlining the experience and through checking for previous experience with filing complaints (McCollough et al., 2000). To guarantee overall comprehension of the surveys even more, a pilot test took place with some close relatives that knew only a little about the research. This was done to check for any wording problems or other errors within the surveys.
Procedures
When pilot testing was finished and the surveys were fintuned, people were invited via e-mail to fill out the surveys. E-e-mail addresses were gathered in the online network through Facebook, LinkedIn and an already existing e-mail contact list. People in the offline network (neighbors and other nearby living acquaintances) were approached offline. People were politely and briefly asked if they wanted to help through participating in a short Dutch internet survey. The used invitational e-mail and reminder e-mail can be found in Appendix C. When people wanted to help, they could click on the Qualtrics link that was sent to them. All invitees were allocated randomly over the four surveys (links) by randomly splitting all e-mail addresses in four equal groups. In addition, all invitees were asked to help reach a few more participants using their own online and offline networks so that a broader group of people would be reached. Some participants (parents, brothers, sisters and close friends) helped to gather even more than a few participants by spreading the four surveys over a large part of their network (work environment and clubs). Instructions were provided to these few helping participants to make sure that the participants they reached, would be fairly divided between the four surveys.
25
When a participant started off with one of the survey versions, first the complaint experience was described. The used experience can be found in Appendix D. The complaint was about white spots on new bought leather shoes and did not vary over the four e-mails. At the same time the e-mail was shown. After reading the e-mail, the participant could click a button to proceed and had to answer the questions, that would appear one by one, while the e-mail remained on the screen. The participant could not go back after proceeding to the next question. After answering seven questions like this, the first three being about complaint satisfaction because this is the dependent variable and the last four being about GSaT, the e-mail disappeared. The e-e-mail disappearing was important because people should not be able to deliberately count mistakes considering the last question in the survey asking about the total of mistakes in the e-mail. Now the final five questions were asked, the first four being control variables (Age, Gender, Education and Prior Experience) and the last being about the total number of mistakes in the e-mail.
Measures
The items used in the surveys to define the dependent and independent variable were seven point Likert scale items that ranged from ‘Strongly disagree’ (1) to ‘Strongly agree’ (7). The seven point scale delivers more response options compared to the five point scale, which means that participants are not forced to choose an option that not fits their ‘true’ answer. The Qualtrics Survey Library was used to look for high quality examples of surveys and items. Through Qualtrics, items that measure the control variables ‘Gender’, ‘Age’, ‘Education’ and ‘Experience’ were found. They are used because it has been suggested that demographics like gender, age and education influence post-complaint behavior and these variables are therefore of influence on complaint satisfaction (Roschk et al., 2013). In general it can be said that men
26
and women differ and therefore they are likely to respond differently. The same goes for different age groups. People might develop a (different) vision on how companies should treat them or they may not care to complain anymore because of their financial and/or health situation (Roschk et al., 2013). Education can be of influence as well because higher educated people overall have better writing and reading skills and might therefore be more aware of mistakes. Of course prior experience with complaining can be of influence because people will know if filing a complaint works and what they can expect from a company to satisfy them.
In comparison to the control variables, peer-reviewed articles used in the literature review provided high quality survey questions for the dependent variable ‘Complaint
Satisfaction’. ‘Complaint Satisfaction’ is the degree to which people feel satisfied after the e-mail response and represents a continuous variable although it is measured as an ordinal variable through three Likert items. This research used items that effectively measured ‘Complaint Satisfaction’ in previous studies. However, the measure names deviate a little from the currently used measure names, but the content of the items is exactly the same. Tax et al. (1998); Wendel et al. (2011). used items like ‘I was not happy with how the organization
handled my complaint’, ‘I was satisfied with the complaint handling of the company’, ‘I had a positive experience when complaining to the company’ and ‘I am not satisfied with the
handling of my complaint‘ to measure ‘Complaint Satisfaction’. GSaT was naturally
measured as a categorical variable because there were four different groups that would provide an answer to H1. A separate variable named ‘SurveyNr’ was therefore created after all the data was collected to represent GSaT in this research. However, for better
understanding of the customer’s point of view towards the quality of grammar errors, spelling errors and typos in e-mail communication and for future research purpose, which will be explained later, GSaT was also represented as a continuous variable albeit that it was
27
measured as an ordinal variable through four Likert items. There were no validated items available from previous studies to measure the point of view of customers towards the quality, so new questions had to be thought of to create appropriate items to measure this continuous variable. The created items were: ‘I am satisfied with the quality of the language in this
e-mail’, ‘This e-mail is given the language, well written’, ‘This e-mail contains many mistakes’
and ‘I would be satisfied about this e-mail when it would contain less mistakes’. Of these items, the continuous variable ‘GSaTTotaal’ was computed by adding up the four Likert item scores.
To determine correlation of the four ‘GSaT’ items, a factor analysis was necessary. Especially because ‘GSaTTotaal’ is a new measure and used non-validated items. This analysis was not necessary for the three ‘Complaint Satisfaction’ items because these items were validated in previous studies and therefore only the Cronbach’s Alpha value was checked. The four control variables consisted of one item per variable. To check whether the participants had a clue about the amount of mistakes in the e-mail after the e-mail
disappeared, there was a variable named ‘FoutenAantal’. In total, twelve items were used. All items had to be translated to Dutch to make sure that the Dutch participants had perfect understanding of the questions asked. All the items can be found in Appendix E.
Results
The data obtained through the surveys was analyzed using SPSS 20 for Windows. H1 was
tested through an analysis of variance (H0: µ1 = µ2 = ... = µk). The means of the four groups (SurveyNr) were compared to see if differences would be found and prove that the total of grammar errors, spelling errors and typos negatively affect complaint satisfaction. The steps
28
taken in the analysis will be explicated. Also as explained before, the extra continuous variable ‘GSaTTotaal’ was tested for future research purposes.
Analysis steps
Before the data was imported into SPSS, a new variable named ‘SurveyNr’ was created in Excel. This variable made sure the data could be split into four groups. The first step in SPSS was to check the data for missing values. The data contained no missing values within the responses of the 223 participants. This was due to the settings (force an answer) used in Qualtrics. Also, only responses that were fully completed were registered. Recoding counter-indicative items was next. The items ‘CS3’, ‘GSaT3’ and ‘GSaT4’ were recoded into same variables since they were counter-indicative. Scores on these items were reversed (1 became 7, 2 became 6, 3 became 5 etcetera). Counter-indicative items were important because
participants may answer all items positive or negative when they do not pay attention. So they were kept alert and at the same time these items checked if answers were consistent.
For understanding the four groups, important general information (mistakes guess, gender, age, education and previous experience) is shown in Table 1. Frequency tables were used as well to interpret the data.
29
Which survey was participated?
20 fouten 10 fouten 5 fouten 0 fouten
Count Count Count Count
How many mistakes are in the e-mail? 0 1 3 4 21 Circa 5 12 15 35 27 Circa 10 21 24 10 7 Circa 15 9 6 5 1 Circa 20 8 5 2 2 Circa 25 5
What is your gender? Man 34 34 31 40
Vrouw 22 19 25 18
What is your age?
18-25 4 4 14 6
26-35 16 15 18 12
36-45 7 9 14 8
46-55 15 10 7 17
56-65 14 15 3 15
What is your highest completed study? Basisonderwijs 2 Speciaal onderwijs vmbo 4 3 2 5 havo 1 2 5 8 vwo 4 1 mbo 9 20 14 12 hbo 25 15 19 20 wo 9 6 2 4 Master 7 4 8 6 PhD 1 3 2
Have you ever filed a complaint against an organization?
Nee 2 1 2 3
Ja, face-to-face 2 3 1 3
Ja, per brief 4 9 5 5
Ja, per telefoon 9 13 21 18
Ja, per e-mail
Ja, via social media 28 19 24 22
Ja, via meerdere van
bovengenoemde wegen 11 8 3 7
30
The table shows that the participants overall made decent estimates of the amount of mistakes without conscious counting of the mistakes (the e-mail disappeared before the question was asked). Only Survey [1] was rated too positive in comparison to the real amount of mistakes. The average mistakes guess for Survey [1] – Survey [4] was: between 15 and 10,
approximately 10, between 10 and 5 and approximately 5 mistakes. Also important, 139 participants (around 62%) were male and 84 participants were female (around 38%). Survey [3] was closest to having an equal division of males and females (n = 31 and 25) while Survey [4] had the most male participants (n = 40) in comparison to female participants (n = 18). Looking at age, the average age for the total participants (n = 223) was 41,85 years. For the separate groups, the age was around 44 years. Only Survey [3] differed with an average age of 35 years because 32 of 56 participants had an age =< 35. The average for highest completed study was between mbo and hbo. 134 Out of 223 participants (around 60%) answered mbo or hbo. Survey [1] (on average hbo) and Survey [4] (on average mbo) differed most. The last questions shows that participants overall filed complaints in the past through social media (n = 93) and telephone (n = 61). Remarkably, none of the participants only filed a complaint in the past using e-mail communication.
Normality was checked for the ‘Complaint Satisfaction’ construct (DV) consisting of three items that were computed to ‘CSTotaal’ by adding up the three Likert item scores. It was also checked for for the ‘GSaT’ construct ‘GSaTTotaal’ (IV, the continuous one). This was done because the data distribution per item was clearly not normally distributed. All items were platycurtic (values from -1,222 to -1,726), two items (‘CS2’ and ‘GSaT4’) were slightly negatively skewed and one item (‘GSaT3’) was slightly positively skewed. Also, the items could not be transformed to a new variable using the log, square root or some other scale. After adding them together, the data distributions looked more normal. Univariate outliers were not present. The histograms of frequencies showed no isolated cases and the
31
standardized scores of the two constructs were not z > |3|. The Shapiro-Wilk test (p > 0,05) was statistical significant (p = 0,000) for both constructs and therefore the data distributions were not normal. However, histograms of frequencies, normal Q-Q plots and box plots showed that the constructs were only marginally non-normally distributed. Kurtosis and skewness where checked by calculating the 95% confidence intervals through the following formula: ‘Statistic + (1,96 * Std. Error)’ and Statistic – (1,96 * Std. Error)’. If zero was in between the intervals, than the variable’s kurtosis and skewness was acceptable. Skewness indicated that the distributions were normal for the constructs, -0,29 (SE 0,163) for
‘CSTotaal’ and 0,16 (SE 0,163) for ‘GSaTTotaal’. In comparison to the separate ‘CS’ and ‘GSaT’ items, the summed variables were still platycurtic, -1,137 (SE 0,324) for ‘CSTotaal’ and -1,323 (SE 0,324) for ‘GSaTTotaal’. These values are only slightly outside the -1 to +1 borders and are therefore not extreme departures from normality. Because of the mixed signals, transformation to another scale was attempted, but again this was not successful since the data suited none of the scales. The original data therefore were used and the distribution was seen as an approximately normal distribution. Perfect normal distributions are rarely seen. Little deviation from normality should have no real effect on the F statistic. Schmider et al. (2010) acknowledge this with their results that give strong support for the robustness of the ANOVA while non-normally distributed data is used. Because of this robustness, the
ANOVA operates well across different distribution scales and therefore the alternative, a nonparametric ‘Kruskal-Wallis’ test, was not chosen. The platykurtosis in the distributions also had no serious effect because of the large and almost equal group sizes used in this research (N min = 53, N max = 58). This view is supported by Schmider et al. (2010). According to the authors, empirical studies often use designs with groups of 25 persons per group because this size is often recommended as the threshold for robustness. Multivariate
32
normality was not checked because the seven Likert items and thus the constructs that follow from them, were not perfectly normally distributed , making testing not useful.
To check for correlation between ‘CSTotaal’ and ‘GSaTTotaal’, a bivariate Pearson correlation test was conducted. It ρ value of 0,406 was found with α = 0,01. This means that the relationship was positive and between weak (+0,3) and moderate (+0,5). A scatterplot of the aforementioned continuous variables confirmed the relationship. For a more in-depth understanding of the correlation, see Table 2.
Which survey was participated? GSaT1-GSaT4 CS1-CS3
20 fouten N Valid 56 56 Missing 0 0 Mean 9,5893 10,7500 10 fouten N Valid 53 53 Missing 0 0 Mean 11,4340 13,3208 5 fouten N Valid 56 56 Missing 0 0 Mean 16,0714 12,8929 0 fouten N Valid 58 58 Missing 0 0 Mean 21,2241 13,5172
Table 2. Relationship between ‘GSaTTotaal’ and ‘CSTotaal’.
An Exploratory Factor Analysis or in other words a Principal Component Analysis (technically not exactly the same), was conducted for the four GSaT items. This was important because GSaT is a non-validated measure. The Kaiser-Meyer-Olkin measure of sampling adequacy found a 0,747 value which meant that the items scored middling and the sample is adequate (> 0,5). Therefore a factor analysis could be attempted. The Bartlett’s Test of Sphericity found a significant score of p = 0,000 which meant that the strength of the relationship among the items was strong and thus the correlation matrix is not an identity
33
matrix. This underlines the outcome of the previous test. Communalities found loadings of 0,800 for ‘GSaT1’, 0,857 for ‘GSaT2’, 0,809 for ‘GSaT3’ and 0,427 for ‘GSaT4’. Table 4 shows the total variance explained.
Factor Initial Eigenvalues Extraction Sums of Squared Loadings Total % of Variance Cumulative % Total % of Variance Cumulative %
1 2,894 72,353 72,353 2,894 72,353 72,353
2 ,699 17,485 89,839
3 ,281 7,024 96,862
4 ,126 3,138 100,000
Extraction Method: Principal Component Analysis.
Table 3. Total variance explained.
The Scree Plot acknowledged that there was only one item with an Eigenvalue bigger than 1. Looking at the loadings (3 loadings > 0,8), the Eigenvalues of the items (1 factor >
Eigenvalue 1) and at the variance explained (72,353% > 60% of Variance), there was only one factor for the GSaT items.
Construct validity for ‘CSTotaal’ and ‘GSaTTotaal’ was important to make sure the data would provide consistent results. Internal consistency was examined for the two summed variables using Cronbach’s Alpha values. The values for ‘CSTotaal’ and ‘GSaTTotaal’ were 0,857 and 0,870. This means that the constructs scales score very high (> 0,7) and thus are valid. The value for ‘CSTotaal’ could be 0,868 if item ’CS3’ would have been deleted. This was however not a significant change in the total Cronbach’s Alpha value (Δ<.10). The value for ‘GSaTTotaal’ could be 0,915 if item ‘GSaT4’ would have been deleted. This was in comparison to previous construct also not a significant change in the total Cronbach’s Alpha value (Δ<.10) and data is ideally not deleted.
34
Finally, the one-way Analysis of Variances (F test) was performed for ‘CSTotaal’ with factor ‘SurveyNr’ after testing for homogeneity of variances among the four groups because equal variance is an important assumption to perform the F test. Levene’s F test indicated that variances were equal and thus the test was not statistical significant (F3,219 = 1,018, p = 0,386)
with α = 0,05. The ANOVA came up with F3,222 = 3,944 (p = 0,009) which means that the test
was statistically significant (α = 0,05) thus there is a difference between the means of the groups. See Table 4 for the ANOVA statistics.
CS1-CS3 Sum of Squares df Mean Square F Sig.
Between Groups 272,714 3 90,905 3,944 ,009
Within Groups 5047,887 219 23,050
Total 5320,601 222
Table 4. Analysis of Variance.
However, at this point it was not clear where the difference was allocated. Therefore a Tukey post-hoc test was conducted. It was found that the control group, Survey [4] (M = 13,517, SE
= 0,590), statistically differed (p = 0,013) from Survey [1] (M = 10,750, SE = 0,671) with α =
0,05. Therefore H1: the more grammar errors, spelling errors and typos, the lower
complaint satisfaction was supported for one treatment group. In addition, Survey [2] (M = 13,321, SE = 0,698, p = 0,29) also had a statistical different mean in comparison to Survey [1] with α = 0,05. The mean for Survey [3] (M = 12,893, SE = 0,616, p = 0,088) was almost statistical significant different as well in comparison to Survey [1] with α = 0,05. See Table 5 for multiple comparisons.
35
Dependent Variable: CS1-CS3 Tukey HSD
(I) Which survey was participated?
(J) Which survey was participated?
Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound
1 (20 fouten) 10 fouten -2,57075* ,92006 ,029 -4,9527 -,1889 5 fouten -2,14286 ,90731 ,088 -4,4917 ,2060 0 fouten -2,76724* ,89945 ,013 -5,0958 -,4387 2 (10 fouten) 20 fouten 2,57075* ,92006 ,029 ,1889 4,9527 5 fouten ,42790 ,92006 ,967 -1,9540 2,8098 0 fouten -,19649 ,91231 ,996 -2,5583 2,1654 3 (5 fouten) 20 fouten 2,14286 ,90731 ,088 -,2060 4,4917 10 fouten -,42790 ,92006 ,967 -2,8098 1,9540 0 fouten -,62438 ,89945 ,899 -2,9529 1,7042 4 (0 fouten) 20 fouten 2,76724* ,89945 ,013 ,4387 5,0958 10 fouten ,19649 ,91231 ,996 -2,1654 2,5583 5 fouten ,62438 ,89945 ,899 -1,7042 2,9529
*. The mean difference is significant at the 0.05 level.
Table 5. Multiple comparisons.
To check for the effect size of the means difference between Survey [4] and Survey [1] and Survey [1] and Survey [2], the Eta squared calculation was executed. This calculation is generally used to reflect the percentage of dependent variable (CSTotaal) variance explained by the independent variable (SurveyNr). The formula is: Sum of squares between-groups (272,714) / Total sum of squares (5320,601). The resulting η2 value was 0,051. This means that the effect was considered an approximately medium effect (0,06 = medium effect).
Discussion
Now that the answer to H1 is known, important conclusions can be drawn to answer the main research question: What is the impact of the GSaT quality measure, in e-mail
communication, on complaint satisfaction? Limitations in this research are noted as well as interesting future research possibilities.
36 Conclusions
This research found that the total of grammar errors, spelling errors and typos has a significant effect on complaint satisfaction thus H1 was supported. As expected, the biggest difference in complaint satisfaction levels was between the two most extreme groups. The control group, Survey [4], was 2,767 lower in mean score than Survey [1]. This seems to be a small difference and is evident looking at the medium effect size that was obtained through the Eta squared calculation (0,051). But because of the large amount of participants (223), small differences can become statistically significant. When broadening the scope, the types of mistakes (striking or more difficult to see) might have influenced the outcomes as well. There were mistakes in the other two treatment groups and one would expect these means (or at least the mean of Survey [2], because Survey [3] only contained five mistakes) to differ from the Survey [4] mean as well. Survey [1] however had compared to the other surveys the most striking mistakes and that might have been of big influence.
It is however interesting to see that the difference in means between Survey [1] and Survey [2] is statistical significant while the difference in means between Survey [1] and Survey [3] was not statistical significant.. Significant difference in means between treatment groups was not necessarily expected. The gap in the amount of mistakes was smaller between Survey [1] and Survey [2] and one would therefore say that this combination should not be significant when considering that the Survey [1] and Survey [3] combination was not
significant. So when looking purely at the total of mistakes, the significant different mean of Survey [1] versus the Survey [2] mean seems to be a strange outcome considering there is still a relatively large amount of mistakes in the e-mail. Table 2 shows clearly that ‘GSaTTotaal’ did not increase by much when looking at Survey [1] and [2]. Perhaps the participants of Survey [2] did not care that much for the mistakes in contrast to the participants of Survey [3]. In addition, the types of mistakes might again have played a large role here and that could
37
explain the significant and almost significant means difference of Survey [1] in comparison to Survey [2] and Survey [3]. Looking at aforementioned, it is hard to say that the total of
mistakes in e-mail communication was the biggest influence on customer satisfaction levels. Types of mistakes seem to matter as well. Therefore the amount of mistakes for complaint satisfaction levels to rise cannot be given. The fact remains that there is a significant impact of GSaT on complaint satisfaction. Therefore writing skills training is important for
employees who are not naturally skilled writers and have excellent command of their
language. Zeithaml et al. (1988) underline this by explaining that training in communication skills should affect the employee’s perceived level of confidence or competence. This should lead to better written communication and with that improved complaint satisfaction.
Another interesting result was that participants in Survey [4] thought that there were around five mistakes in the mail. This is a bit peculiar since there were no mistakes in the e-mail for Survey [4]. Also, the guess for total mistakes in the e-e-mail was only spot on for Survey [2]. These outcomes suggest that people often do not see mistakes and/or think there are mistakes while there are not. Looking at individual response, it becomes clear that a lot of participants do not have the proper ‘written language level’ which is needed to spot mistakes and/or determine if text is well written. This idea is clearly in favour of the types of mistakes made being a big influence.
Also, none of the participants answered to have filed a complaint via e-mail in the past. However, it can still be true that some of the participants did this because of the option that stated ‘Yes, through one or more of the above paths’. Especially social media and
telephone were popular media for filing a complaint in the past and therefore probably also in the future. It remains peculiar to see that e-mail was not popular looking at the different literature and the rate at which it is used. This is possibly due to the average age of the participants.
38 Limitations
There were a few limitations that should be enlightened. The first being the validity of item ‘GSaT4’. After some in-depth analysis it looked like most participants answered this question as if the question was posed positively. Of course the Survey that they were in was checked as well (look at the answers given for GSaT1-3) to back this assumption. Somehow this
happened to many of the participants. Rereading it now, it can be understood that participants found this question hard. However, this did not come forward in the pilot test. If the
assumption is correct that the question did not measure what was intended for all participants, than the higher Cronbach Alpha’s score when the item would be deleted would be
understandable. Also the lower loading score, compared to the other three GSaT items, in the factor analysis would be clear.
Officially, Likert items are not meant for calculating mean scores. Likert items
generate ordinal data while the data is often interpreted as interval data. This is however, done very often and is generally accepted within research. Therefore, the assumption that was made in this research, is that the psychological distance between each answer on the Likert scale (for example, agree – totally agree) is the same as any two other answers that are next to each other on the scale. It was not explained to the participants that each distance should be
considered the same. However, looking at the answers given, it seems that the participants interpreted the scale as was intended up front.
Strictly speaking, interpunction is important when written communication is the subject. Interpunction errors however, were not tested because the assumption was made that it would need participants that are highly skilled when it comes to written language, to uncover these mistakes.
39
Considering culture, Dutch participants will probably differ significantly compared to people from other countries and thus their attitudes towards GSaT within e-mail
communication by complaint management. Generalization of the participants to the whole population of e-mail users worldwide is therefore not desirable.
Future research
Findings enabled new interesting ideas to come to mind. One is checking whether participants have the proper ‘written language level’ to spot mistakes and/or determine if text is well written. Many of the participants could not exactly guess how many mistakes were in the e-mails. To be absolutely sure about the ‘written language level’ of participants, the question about the amount of errors in the e-mail could be asked while the e-mail would stay on the screen. People would have the opportunity to count. In addition, the question could be open without having the choice in several answers beforehand. Participants would have to fill in the amount of errors themselves and should give an even better estimate of their level.
Another idea would be to rewrite the Survey [1] e-mail so that most striking mistakes in it will be replaced with less striking mistakes. This way it can become clear whether the amount of errors or the types of mistake has the bigger influence on customer satisfaction.
The main future research possibility could lie in adding theory to the conceptual model
to further explore the subject. A continuous variable (‘GSaTTotaal’) was already explored to see if there would be a relationship with complaint satisfaction. Since there is one according to the positive correlation found, the conceptual model could be made more complex. Originators of the five P’s standards model (Murphy & Gomes (2003); Murphy & Tan (2003)), did not integrate outcomes in their five P’s model to be of influence on how to respond to e-mail communication. But since customers complain for a different outcome, this
