Effects of Pictographs and Quoting on Flaming in Computer Media

Phillip A. Thompsen and Davis A. Foulger

Abstract

This study examines the nature of flaming (a form of hostile communicative behavior) in computer media by exploring, in the context of five escalating levels of argumentative message, the effects of the pictographs and quotations (which are commonly found in electronic messages) on the perception of flaming in messages. It is hypothesized, based on observations in the literature, that the presence of pictographs should reduce perceptions of flaming while the use of quoting should increase perceptions of flaming. While no specific hypotheses are offered relative to message levels, it is expected that the perception of flaming will increase as the intensity of messages progress from "shows divergence" through "shows disagreement", "shows tension", and "shows antagonism" to "shows antagonism using profanity".

This last expectation is confirmed, with significant changes in the perception of flaming associated with each message level. No perception of flaming is apparent for messages that simply show divergence. Little perception of flaming is apparent for messages that show disagreement. Moderate flaming is perceived in messages that show tension. Strong perception of flaming is found for messages that show antagonism. Obscenity has a small but significant effect in increasing the perception of flaming. It can be concluded, based on these results, that a message will be perceived as a flame when it expresses antagonism toward another interactant, regardless of the use of profanity.

While a significant moderating effect is observed in messages that included pictographs, the effect is not as strong for true flames (e.g. messages that show antagonism) as it is for messages that simply show tension. Nonetheless, pictographs appear to have a substantial effect in moderating the perception of messages as flames. The results associated with quotations are ambiguous. Although no significant effect was found in the primary analysis, a secondary analysis finds a potentially complex relationship of quoting to messages in which quotation moderated the perceptions of non-antagonistic messages but amplified the perceptions of true flames.

Introduction

As the use of computer technology for human communication becomes more widespread, computer-mediated communication is receiving an increasing amount of attention from the scholarly community. A major research concern has been the potential effects of the computer medium on how humans communicate with one another. Studies have examined the effects of computer-mediated communication in at least three communication contexts: small group communication (Hiltz, Johnson, & Turoff, 1986; Kiesler, Siegel, & McGuire, 1984; Lea & Spears, 1991; Siegel, Dubrovsky, Kiesler, & McGuire, 1986; Spears, Lea, & Lee, 1990), organizational communication (Huff, Sproull, & Kiesler, 1989; Komsky, 1991; Rice & Shook, 1990; Rubinyi, 1989; Sherblom, 1988) and interpersonal communication (Blackman & Theodore Clevenger, 1990; Holmes & Berquist, 1990; Rice & Love, 1987; Shamp, 1989; Walther, 1992). Many of these studies have found evidence suggesting computer-mediated communication may affect the socio-emotional content of communication, often in socially undesirable ways.

The present study explores what has been suggested (Keisler et al, 1984) as a distinctive-to-computer-media form of such undesirable socio-emotional content: flaming. Flaming describes a form of hostile verbal behavior in computer-mediated communication. This exploration examines the effects of two stylistic conventions routinely associated with computer media on the perceived negative socio-emotional content of a computer-mediated discussion which is conducted at five levels of escalating disagreement.

A two-factor experimental design examines the effects of pictographs and the quoting style of response on perceptions of "flaming." Pictographs are strings of typographic symbols often used to express emotion or as surrogates for nonverbal communication; for example, :-) is a commonly-used pictograph representing a "smiley face" on its side. Quoting is a style of response in which the author of an electronic mail message incorporates full or partial quotes from a prior message in responding to that message; typically, these quotes are prefaced by the greater than symbol (>), and may be interspersed with comments in response to the quoted topic.

This two-factor design is applied to messages at five levels of escalating disagreement. Four of the five message levels, shows divergence, shows disagreement, shows tension, and shows antagonism, correspond to Bales' interaction process categories. In a fifth level, "shows antagonism" is reinforced through the use of mild profanity. It is hoped that this escalation will inform the proper definition of when a message becomes a "flame".

When is a message a flame

Among the phenomena that have been observed in computer-mediated communication, one of the most widely commented on, and perhaps least understood, is what is popularly known as flaming. Although the term apparently evolved from the "hacker" computer subculture, flaming has been used as a research variable in a number of studies (Lea, O'Shea, Fung, & Spears, in press). Flaming has not, however, been consistently operationalized across these studies (Anderson & Walther, 1992). The definitions of flaming have been offered in both the scholarly and popular press include:

• the hostile expression of strong emotions and feelings (Lea, et al., in press);
• the posting of passionate tirades (Levy, 1992, p. 65);
• insults, swearing, and hostile, intense language (Walther, 1992);
• the practice of expressing oneself more strongly on the computer than one would in other communication settings (Kiesler, et al., 1984, p. 1130);
• speaking incessantly, hurling insults, and using profanity (Baron, 1984, p. 130);
• speaking rabidly or incessantly on an uninteresting topic or with a patently ridiculous attitude (Steele, Woods, Finkel, Crispin, Stallman, & Goodfellow, 1983, p. 65).

These definitions are all focused on attributes of messages. Indeed, they are consistent in their suggestion that the term "flame" describes a form of extreme messaging behavior. The range of the definitions is problematic, however, with strong emotions, hostility, passion, insulting messages, intensity, profanity, ridiculousness, message volume (interpreted as either excessively long messages or excessive numbers of messages), and irrelevance all cited as being characteristic of flames. It seems clear that we will not obtain a clean operationalization of flaming in the union of these definitions.

From a communication perspective, it may be more appropriate to conceive of flaming as a perceived quality of a communication interaction, since flaming emerges when a reader attributes this quality to a message or series of messages. In other words, a flame is not a flame until someone considers it a flame. What distinguishes a message as a flame, then, is less a characteristic of a message than it is an emerging quality attributed to a sequence of messages by human actors involved in an interpretive process of meaning creation (Anderson & Meyer, 1988; Leeds-Hurwitz, 1992). From such a perspective, an operational definition of flaming would focus on the perception of messages, specifically, the extent to which a person considers a sequence of message to constitute an episode of flaming.

This perspective suggests that one good way to explore what people mean when they use the term flaming is to have them rate a variety of messages that entail varying levels of extremity of expression. Several of the categories associated with Bales' interaction process analysis (e.g. shows divergence, shows disagreement, shows tension, and shows antagonism) form a continuum of such extremity that might prove useful in arriving at a perceptual definition of flaming. Some of the specific characteristics that appear across definitions, including the use of profanity, might be usefully manipulated within these categories.

Experimental Factors: Pictographs and Quoting

Sometimes referred to as "emoticons," (Blackman & Clevenger, 1990) or "relational icons" (Asteroff, 1987), pictographs have been examined for their utility as nonverbal surrogates in computer-mediated communication. A commonly cited limitation of the text-based mode of communication used in electronic mail and computer conferences is the inability to use nonverbal communication to convey social context cues (Kiesler, et al., 1984; Sproull & Kiesler, 1986). Blackman and Clevenger (1990) suggest that electronic mail users employ pictographs to overcome this limitation: "Lacking the rich multichannel sensory input that characterizes face-to-face communication, they invent and use symbol strings that substitute for the audible, visible and tactile elements of interpersonal communication" (p. 15). Pictographs may also help convey the intended tone of a message, providing a signal to the reader that a comment is not meant to be taken too seriously. Some observers (Foulger, 1991; Goode & Johnson, 1991; Turner, 1988) have suggested that the use of pictographs can be an effective deterrent to flaming, or at least slow the growth of a simple divergence of opinion into a polarized debate; this study attempts to experimentally test this suggestion.

There is some disagreement as to the potential effects of the quoting style of response. Goode and Johnson (1991) suggest that including the salient part of a previous message in a response to that message is a useful convention; otherwise, the asynchronous nature of electronic mail discussion can make it difficult to follow a discussion "thread" among the clutter of many simultaneous conversations. This problem has also been noted in synchronous computer conferencing (Holmes, 1987). The convention of using the greater than symbol to indicate quoted material appears to be widespread, and many electronic mail systems automatically insert these symbols as part of a reply command. Yet despite the tacit sanctioning of the convention by its inclusion in electronic mail software, some have suggested that it should be discouraged. Foulger (1991) argues that this style of response, which he refers to as the "quote/antiquote" style, often causes messages to "appear confrontational even when they are not intended to be" (p. 200). He also suggests that the quoting style of response may be "one of the key instigating elements" of flaming (p. 187), a notion that is explored experimentally in the present study.

Pictographs and quoting, then, were the two independent variables manipulated in this experiment, in what is commonly referred to as a "judgment study" (Aronson, Brewer, & Carlsmith, 1985). The manipulation of these variables was straightforward, as the essentially typographic characteristics of the two factors could simply be added to the stimulus material, an episode of flaming consisting of a sequence of 11 electronic mail messages (See appendix). To enhance the face validity of the stimulus material, the experimental manipulations were conducted on a subset of the message sequence, with pictographs added to 6 messages, and the quoting style added to 6 messages.

Hypotheses

Since pictographs may serve as nonverbal surrogates, suggestive of facial expression, they may add a paralinguistic component to a message. Pictographs may thus enhance the exchange of social information, by providing additional social cues beyond what is found in the text of a message. If flaming is a result of reduced social cues, as suggested by Sproull and Kiesler (1991), then pictographs, by increasing social cues, should have a modifying influence. Such an influence could be revealed by less intense perceptions of flaming. Thus,

• H1: When pictographs are included in the sequence of messages, subjects will assess flaming as less intense than when pictographs are not included.

The quoting style of response may be, as Foulger (1991) suggests, a significant "instigating element" in the process of flaming. If so, the hypothesized effect of quoting in this experiment would be the converse of the first hypothesis:

• H2: When quoting is included in the sequence of messages, subjects will assess flaming as more intense than when quoting is not included.

Given the range of definitions of flaming, it is currently difficult to exactly state when a message is or is not a flame. Given, however, the consistent definition of flaming as involving extreme messaging behavior, is is reasonable to expect that the perception of flaming in a message will increase as a message becomes more extreme. Several escalating categories of Bales' interaction process analysis, including shows divergence, shows disagreement, shows tension, and shows antagonism, offer the possibility of manipulating the extremity of messages in a systematic and escalating fashion. While one is hesitant to hypothesize exactly when a message becomes a flame, it is reasonable to also ask the research question:

• RQ: At what point in the escalation of argument in a message can we conclude that a message has become a flame?

Method

Subjects

One hundred sixty-four subjects voluntarily participated in the experiment. Invitations to participate were sent via electronic mail to a sample of 900 Bitnet and Internet electronic mail addresses. This sample was randomly drawn from the publicly available subscriber lists of 15 randomly selected electronic mail discussion lists from the Bitnet "List of Lists," a publicly available list of electronic mail discussion lists. Electronic mail users who wish to participate in an asynchronous discussion on a particular topic of interest subscribe to a discussion list; they are the electronic mail equivalent of a computer conference or discussion forum. The 15 randomly selected discussion lists were drawn from a pool of 3,136 lists. Of the 900 invitations, 91 were returned as undeliverable, resulting in a pool of 809 successfully delivered invitations. However, it was not clear how many of these invitations were delivered to an active electronic mail account, and how many of those were actually read. The overall response rate was thus 20.3%, although the rate of response from those who may have read the invitation is likely to be much higher. As a check on the diversity of the sample, subjects were asked at the conclusion of the experiment to indicate their state or country of residence. Subjects represented 23 different nations, with 68.3% of the subjects from the United States (37 different states) and 18.3% from Europe. The average subject was 34 years old.

Apparatus

An interactive computer program was developed for administering the experiment. The program was written in the VMS Digital Command Language (Digital Equipment Corporation, 1988) and ran on a Digital Equipment VAX main frame computer at the computer center of a major North American university. Subjects accessed this program through the BITNET or Internet computer networks by using "remote log-in" telecommunications software (such as TELNET). The program randomly assigned subjects to one of the four experimental groups (control, pictographs, quoting, and both). The program presented the stimulus material (a sequence of electronic mail messages), administered the perception of flaming semantic differential scales, and asked subjects a series of related questions.

Procedure

Stimulus material. Each subject was presented with an episode of flaming, consisting of a sequence of eleven electronic mail messages, in which the presence of pictographs and quoting was manipulated with five levels of message intensity (see appendix for the specific messages). The flaming episode was presented to subjects as a small portion of an "ongoing, organization-wide group discussion, conducted by way of electronic mail." The flame evolved in a sequence of increasingly hostile messages from two primary participants, "Snow Pro" and "Dr. Ski," who responded to a query from "A Total Novice" about where to learn to ski. Pseudonyms were used to identify the participants in order to suppress gender attribution effects (Robinson-Stavely & Cooper, 1990).

The escalation of message intensity was symmetrical, with each participant offering one message at a given level before proceeding to the next level. Dr. Ski initiated each escalation. "Snow Pro" accepted each escalation bid with a message in kind. The escalation proceed from "shows divergence" (messages 2 and 3), through "shows disagreement" (messages 4 and 5), "shows tension" (messages 6 and 7), and "shows antagonism" (messages 8 and 9), to "shows antagonism using mild profanity" (messages 10 and 11).

The manipulation of pictographs and quoting proceeded within these couplets. Quotation was manipulated in messages 2, 3, 4, 5, 8, and 9. Pictographs were manipulated in messages 6, 7, 8, 9, 10, and 11.

Measuring perception of flaming. Subjects were asked if they were familiar with the term flaming, and if so, to assess the extent to which the sequence of messages represented an episode of flaming (using a single seven-point semantic differential scale anchored by "flaming" and "not flaming"). Subjects not familiar with the term flaming were provided with a brief definition ("the heated exchange of messages expressing hostility") and then asked to assess the extent to which the sequence of messages represented an episode of flaming. Subjects were then presented with each of the 11 messages a second time, and after each message, asked to indicate the extent to which the sequence of messages, up to that point, represented an episode of flaming (again using the single seven-point semantic differential scale).

Manipulation checks. As a manipulation check of the experimental factors, subjects in the pictograph conditions were asked if they had seen pictographs before (85.4% said they had), and if they knew what they represented (81.7% said they did). Similarly, subjects in the quoting conditions were asked if they had seen this style of response before (94.8% said they had), and if they knew what it represented (90.9% said they did). Finally, as a check on familiarity with electronic mail, subjects were asked to indicate the number of electronic mail messages they read daily (the average was 41.4 messages) and the percentage of these messages that included flaming (the average was 9.3%).

Analysis. The data analysis was conducted using stepwise multiple regression with the semantic differential perception of flaming as the dependent variable. Each message viewed by each subject was treated as a case, with the 161 valid subjects expanding to 1771 cases. The primary analysis involved seven main effect variables and six interaction variables:

DIVERGE, ARGUMENT, TENSION, INSULT, and PROFANE

Dummy indicators of the five escalating levels of argument. These variables nest, with each level encompassing the higher levels. Hence DIVERGE encompasses messages 2 through 11, ARGUMENT encompasses messages 4 through 11, TENSION encompasses messages 6 through 11, INSULT (antagonism) encompasses messages 8 through 11, and PROFANE encompasses messages 10 and 11.

PICXARG, PICXINS, and PICXPRO

Three interaction effects of the various levels of interaction and the presence of pictographs in a particular subject's messages. PICXARG is the interaction of pictographs and argument. PICXINS is the interaction of pictographs and insult (antagonism). PICXPRO is the interaction of pictographs and the use of profanity. Limitations in the experimental design (already described) made both the main effect for pictographs and other theoretically possible interaction effects duplicative of these three effects. Hence neither the main effect nor other theoretically possible interactions were tested.

QUOXDIV, QUOXARG, and QUOXINS

Three interaction effects of the various levels of interaction and the presence of quotations in a particular subject's messages. QUOXDIV is the interaction of quotes and shows divergence. QUOXARG is the interaction of quotes and argument. QUOXINS is the interaction of quotes and insult (antagonism). Limitations in the experimental design (already described) made both the main effect for quotes and other theoretically possible interaction effects duplicative of these effects. Hence neither the main effect nor the other theoretically possible interactions were tested.

DRSKI and SNOWPRO

The two principle interactants in the message series. DRSKI is "Dr. Ski". SNOWPRO is "Snow Pro".

All of these independent measures were coded as dummy variables. The use of multiple regression, in this case, allows us to explore specific variable relationships that would be assessed equivalently, but somewhat differently, in an analysis of variance. Indeed, the results of the essential two by two design duplicate the results of a MANOVA (performed using the 161 subjects as cases) which is not reported here. The preference for multiple regression over MANOVA in this study reflects the increased flexibility that regression allows in setting up the independent variables for analysis. In particular, the analysis of the five levels of argument would take a different and omewhat less clearly interpretable form in an ANOVA or MANOVA than is possible in a well constructed multiple regression analysis.

A secondary analyses (also a stepwise multiple regression) is also reported here. It omits the TENSION variable with interesting results.

Results

The results of the primary analysis are reported in table 1. This analysis (F=732.99, p<.0001) accounts for 74% of the variance in the perception of flaming as measured in this manipulation. Because all the independent variables in this analysis are coded as dummy variables, the unstandardized beta weights (B) can be translated directly into movement on the seven point measure of perception of flaming. A B of 1 for a given variable, for instance, means the presence of that variable translates to an average 1 point increment in the perception of flaming. A B of -1, by contrast, translates to a one point decrement in the perception of flaming.

It should be expected, given the research question, hypotheses, and method, that:

• escalating intensity of argument as manipulated in the five message levels should result in a positive B in which the magnitude of B represents an increment on the semantic differential that measures perception of flaming.
• the presence of pictographs should result in a negative B that represents a decrement on the semantic differential that measures perception of flaming.
• the presence of quotations should result in a positive B that represents an increment on the semantic differential that measures perception of flaming.

When is a message a flame?

These expectations are, for the most part, satisfied. While no specific hypotheses were offered relative to message levels, it was expected that the perception of flaming would increase as the intensity of messages progressed from shows divergence to shows antagonism using profanity. This expectation is supported through a series of increments in the perception of flaming. There was effectively no perception of flaming apparent for messages that simply showed divergence (DIVERGE does not enter the equation; t=1.091, p=.2754). Indeed, both the means of messages two and three (1.13 and 1.16 respectively) and the constant of the regression equation (1.02) fall at the non-flaming extreme of the seven-point semantic differential scale used in this study. This result should not be surprising given the content of messages two and three. Each simply offers information in response to the initial query. Simple divergence of opinion does not, then, constitute flaming.

Little perception of flaming was apparent for messages that showed disagreement. The unstandardized beta weight associated with ARGUMENT is .79 (t=8.605, p<.0001); indicating a very small increment in the perception of flaming relative to the regression constant. The mean perception of flaming for messages four and five (1.71 and 2.23 respectively) are consistent with this very moderate (less than one point) increase in the semantic differential perception of flaming. The simple act of showing disagreement cannot, it would appear, be construed as flaming by any reasonable perceptual definition.

The effect of the addition of tension to a message is much more substantial. Indeed, TENSION, as manipulated here, adds the largest overall average increment to the perception of flaming in messages. Its unstandardized beta (B=2.24, t=18.457, p<.0001), when added to the effect of ARGUMENT, adds 3 full points to the semantic differential measure of the perception of flaming, establishing messages that entail tension at the midpoint of its seven point scale. The means for messages 6 and 7 (3.6 and 4.12 respectively) are, once again, consistent with this expectation. Given the standard deviation associated with these messages (1.85 and 1.92 respectively), one can reasonably expect that at least some messages that entail tension will be regarded as flames. One cannot, however, conclude that tension, alone, can be used to routinely establish messages as flames.

It is clear, however, that there is a strong perception of flaming associated with messages that showed antagonism. The unstandardized beta increment associated with the INSULT variable is 1.75 (t=13.376, p<.0001). When added to the terms associated with ARGUMENT and TENSION, this results in expected semantic differential perceptions of flaming in the range of 5.75; a value that is strongly skewed toward the flaming end of the scale. The means for messages 8 and 9 (5.52 and 6.09 respectively) bracket this expected value. The use of profanity further strengthens the perception of flaming in messages that show antagonism (B=.69, t=6.899, p<.0001), incrementing the semantic differential, on average, by nearly another three-quarters of a point. This increment is small compared with the increment associated with antagonism, but clearly justifies the association of profanity with flames as an intensifying element.

It can be concluded, based on these results, that a message will be perceived as a "flame" when it expresses antagonism toward another interactant. Profanity will increase this perception that a message is a flame, but profanity is not "necessary" to flaming.

Pictographs and flaming

The hypothesis that the presence of pictographs would reduce perceptions of the intensity of the perception of flaming was strongly supported (PICXARG) in the regression. The interaction of pictographs and argument results in a reduction of almost three quarters of a point in the perception of flaming on the semantic differential (B=-.71, t=5.090, p<.0001), enough to move the average evaluation of a message that entails argument but not antagonism below the scales mid-point. This effect is weakened where there is antagonism, however, with the interaction of pictographs and antagonism (PICXINS) incrementing the semantic differential by .37 (t=2.02, p=.0278). This interaction, when taking together with the unstandardized beta of PICXARG, translates to a pictograph induced decrement in antagonistic messages of only -.34.

This result is an interesting one, as it indicates that the effect of introducing antagonism to a message, in addition to its effect in incrementing the perception of flaming, reduces the overall credibility of pictographic message modifiers.

Quotations and flaming

Although the interaction of quotation and antagonism (QUOXINS) exhibits a trend in the expected direction (B=.02, t=1.435, p=.15), the hypothesis that the use of quotation of the content of prior messages would increase perceptions of the intensity of flaming is not supported in this analysis. Interestingly, however, a single change in the analysis changes this situation substantially. Table 2 omits the main effect of shows TENSION. Two interactions with quotation move in to fill the void, with the analysis otherwise relatively indifferent from the primary analysis. The first of these two interactions, QUOXARG (quotation with argument) runs counter to the hypothesized expectation. With a standardized beta of -1.25 (t=9.38, p<.0001), this interaction indicates that quotation can have a powerful effect, in normal argument, in the direction of reducing the perception of flaming. The second of these interactions, QUOXINS (quotation with antagonism) operates, however, in the hypothesized direction, incrementing the perception of flaming by 1.45 (t=7.303, p<.0001). Taken together, these measures indicate a possible increment in the perception of flaming by .20 when quotation is used.

One hesitates to make too much of these results given that "shows TENSION" washes out the entire effect to a trend. Weaknesses in the experimental manipulation can easily be credited with either negating the expected effect of quotations or creating a spurious effect where there was none. Still, given the trend in the primary analysis and the interesting double jointed results in the secondary analysis, one cannot dismiss the possibility that quotation does have value in electronic messages that is rather more complex than was hypothesized. It would appear that, in the presence of normal argument, quotation has the desirable effect of creating a precise reference for ones comments. Quotation, in this context, gives the appearance of given a considered and highly specific response to prior arguments. This effect appears to be cancelled entirely (plus a little) in the face of an antagonistic message. Here the use of quotations in ones reply may be regarded as a cynical attempt to twist a writer's words against them.

An order effect?

The final significant measure in the primary analysis is DRSKI, the messages associated with Dr. Ski. While this result is not associated with either the hypotheses or the research question, it does bear comment, as it may be indicative of an order effect that may be worth exploring. Dr. Ski always offers the second response in the message series. In effect, his role at each junction of this manipulation has been to accept an escalation that was initiated by "Snow Pro". While it is possible that this result is simply an order effect artifact, it raises the possibility that readers evaluate initiators of an escalation in an argument differently than they evaluate those who accept the escalation. It appears possible that the initiator of an escalation is not evaluated as negatively as those that subsequently accept the escalation.

Discussion

This study suffers from a significant design flaw. It was designed as pure two by two manipulation appropriate for multiple analysis of variance. This kind of design, common in communication research and the social sciences in general, can generally be faulted for forcing an oblique and often continuous world into an orthagonal and often dichotomous box. For most experiments the decision to operate this way is easy to justify. In this experiment, it creates a number of problems that would have been obviated by a more flexible design. The problem, in this case, is that quotations and pictographs were assigned to each group of subjects in an identical way with neither quotations nor pictographs evident for all levels of messages. There is, for instance, no testing of pictographs for either the "shows divergence" or "shows disagreement" message levels. There is, moreover, no testing of quotations for either the "uses profanity" or "shows tension" message levels.

A better design would have assigned pictographs and quotations to all or most messages (and certainly assigned each to at least one message within each level) and then used pictographs and quotations randomly across all or most subjects, with the condition associated with each message recorded as a part of the output data. This would have better approximated the real world where the use of pictographs and quotations varies substantially, both from message to message and, in broad and highly noticeable ways, across networks and computer conferencing software front ends. This would have made for a richer and more complete analysis that does not have some of the obvious holes associated with this one.

A design like this would have enhanced out understanding of the use of quotations in the presence of profanity, the use of pictographs in the presence of argument, and, perhaps, the effects of quotation in the presence of the TENSION. It would have also allowed for a clean examination of the lag effects of quotation or pictographs in prior messages on the perception of subsequent messages. These faults don't diminish the value of what has been learned in this study, but they do suggest that we might have learned more if the study had been designed around a multiple regression analysis from the start.

Lessons for future studies

There remains a substantial richness in this study and its results, however, that have immediate implications to our understanding of flaming. Indeed, a large number of future research directions are suggested by these results:

• A flame can be reasonably defined, in the wake of this study, as a message that is antagonistic to the writer of another message. One wonders, however, what the effects would be if the focus of one's antagonism was directed at the prior message rather than the writer, at the idea discussed in the prior message rather than the writer or the message, or a third party referenced in the prior message. One suspects, in each case, that the indirection would result in a diminishing of the evaluation of flaming.
• It appears that while profanity is not necessary to defining a message as a flame, that it does significantly intensify the effect associated with antagonistic messages. One wonders, however, how profanity operates in the context of less extreme messages. What effect does profanity have in messages that simply show divergence (e.g. "Brighton is a damn good place to learn to ski.") or other non-antagonistic levels of argument. One suspects that the use of profanity will increase the perception of flaming.
• While the hypothesis associated with quotation is not supported in this study, there are clear indications of a rather more complex effect associated with quotation that is consistent with, but more subtle than was expected. This more complex effect warrants further study in an experiment that explores quotation for all levels of messages and makes more pointed use of quotations in a manner that actually twists the words of the quoted writer.
• This studies use of a pure symmetrical escalation through five levels of argument describes only one of many possible paths a discussion can take. It would be interesting to see what happens in the face of a refusal to accept a bid of escalation in such interaction. How are the escalators messages regarded in the wake of a non-escalation or non-escalating defense? How are the non-escalating messages regarded under such circumstances? Is the non-escalating message painted as more of a flame in the wake of an escalating message? Is the escalating message painted as more of a flame in the wake of a pointedly non-escalating defense?
• In a similar vein, how are messages regarded when the escalation is discontinuous (e.g. where a respondent goes from argument to antagonism in a single jump?
• What happens to the evaluations when the initial questioner interrupts the argument with a "Thanks for the information. It really helped" note and the argument continues anyway?
• What happens when a party (one of the participants or a third party) jumps in with meta-discussion along of the lines of "enough already"? How might subsequent meta-discussive flames be evaluated?
• What happens when a third party to an argument jumps in precipitously with an antagonistic message?
• Only one pictograph is used in this study. What differences in effect are associated with other commonly used pictographs (e.g. ;-), :-/, :-(, etc).
• The messages in this study are all relatively short. What effect would longer messages have on the perception of flaming?

Is flaming more general than we have thought?

There are clearly a wide variety of other studies that can be usefully done using this model. Interestingly, this study suggests that such studies of computer-mediated interaction may be more general to other communication contexts than has been indicated by prior research. It may be that Keisler et al (1984) did a considerable disservice to the generality of flaming when they described it as being distinctive to the computer mediated environment. Communication acts akin to flaming can be identified in many media, including simple face to face interaction (a boss reaming out an employee; a spouse dumping on a husband or wife), letters to the editor, political campaign speeches, shock talk broadcast radio, and other contexts. Indeed, it is relatively easy to identify contexts in which flame-like behavior is more the rule than the exception, including some construction sites and (to all appearances based on viewing such meetings on C-Span) most televised meetings of the British and Canadian House of Commons.

The relatively simple operationalization of flaming that emerges from this study (a flame is a message that is antagonistic toward a previous message) raises questions of just how distinctive flaming is. Antagonistic messages occur in many contexts, and it is hardly unusual for such messages to be regarded as undesirable by their recipients. The occurrence of such behavior appears to be highly rule-bound. A teenager who hardly utters a profanity-free or non-antagonistic paragraph when with friends may completely avoid such language in home, classroom, and other rule-bound social contexts. Indeed, when that teenager turns into a parent they may not only avoid such behavior altogether, but find the same behavior highly offensive in their own children.

Foulger (1991) has observed this kind of rule variation in computer mediated communications. His initial observation of such variation came in a comparison of Keisler et al's results with the behavior on IBM's internal computer conferencing facility, where flaming is more the exception than the rule and profanity is almost non-existent. Similar variation can be observed elsewhere, however. While there have been discussions on BITNET's COMSERVE in which flaming was more the rule than the exception, other active COMSERVE conferences have seen very little flaming and there has been at least one occasion where concerted community action brought a persistent flamer under control. Similar variation can be observed on USENET, PRODIGY, and CompuServe conferences.

These results and observations through both the necessity and distinctiveness that Keisler et al (1984) associate with flaming and electronic media into serious question. It is not at all clear that flaming is distinctive to computer conferencing. It seems fairly obvious, moreover, that flame-free computer conferences are a reality. Flaming, it would appear, is ultimately a well publicized artifact of a set of limited experimental manipulations that does not generalize to electronic media as broadly as was first thought. This does not make flaming any less interesting as a focus of study. Indeed, it raises the possibility that lessons learned in simple computer-based manipulations may have considerable generality in other media contexts.

Conclusion

This study has examined the nature of flaming in computer media by exploring, in the context of five escalating levels of argumentative message, the effects of the pictographs and quotations that are commonly found in electronic messages on the perception of flaming in messages. It was hypothesized, based on observations in the literature, that the presence of pictographs would reduce perceptions of flaming while the use of quoting would increase perceptions of flaming. While no specific hypotheses were offered relative to message levels, it was expected that the perception of flaming would increase as the intensity of messages progressed from "shows divergence" through "shows disagreement", "shows tension", and "shows antagonism" to "shows antagonism using profanity".

This last expectation was confirmed with significant changes in the perception of flaming associated with each message level. No perception of flaming was apparent for messages that simply showed divergence. Little perception of flaming was apparent for messages that showed disagreement. Moderate flaming was perceived in messages that showed tension. Strong perception of flaming was found for messages that showed antagonism. Obscenity had a small but significant effect in increasing the perception of flaming, but it can be concluded, based on these results, that a message will be perceived as a "flame" when, it expresses antagonism toward another interactant, regardless of the use of profanity.

While a significant moderating effect was observed in messages that included pictographs, the effect was not as strong for true flames (e.g. messages that showed antagonism) as it was for messages that simply showed tension. Nonetheless, pictographs appear to have a substantial effect in moderating the perception of messages as flames. These results suggest that pictographs modify perceptions of flaming, and that they may represent a useful strategy for preventing unintentional outbreaks of flaming. Pictographs seem to alert readers that a message may be taken less seriously, and thus less threatening in terms of flaming potential. In an ongoing discussion where divergence of opinion, disagreement, and tension may already be present, pictographs may help delay the escalation of tensions, and thus reduce the occurrence of flaming.

The results associated with quotations were more mixed. Where the variance associated with showing tension in messages was removed, no significant effects were found for quotations. Where this variance was not removed from the analysis, however, there was a significant effect in the hypothesized direction. Indeed, the analysis revealed a potentially complex relationship of quoting to messages in which quotation moderated the perceptions of non-antagonistic messages but amplified the perceptions of true flames. These mixed results for quoting may be an artifact of the experimental design. Hence, while it cannot be concluded that quoting had the hypothesized effect, neither can the possibility of such an effect be dismissed.

This ambiguous result is only one of a number of research directions suggested by this study, including examinations of the effects of indirectly antagonistic messages, profanity in non-antagonistic contexts, refusals to escalate argument, meta-discussions, interruptions, discontinuous escalations, and other pictographs on the perception of flaming.

References