2. La construcción de una política de la verdad a través de la astronomía
2.2 Ciencia, comunicación y política de la verdad
Together with the questions relating to the model, there were also a number of demographic questions, the results of which are outlined below. One participant who completed the study did not submit demographic data, so resulting in the demographic results being based on 407 participants. Formatting has been completed to ensure that each table remains on one page, for the ease of comprehension.
Sex & Age
93 (22.9%) of the participants were female, with an average age across the sample population of 26.56 years (SD=9.06), with a range between 18 and 68 years old.
Broken down into 10 year age brackets, the sample based on age is shown in Table 5.5.
Table 5.0.5: Age of Participants
Age Group Number of Participants
18 – 24 224 (55.0%)
25 – 34 113 (27.8%)
35 – 44 52 (12.8%)
45 – 54 12 (2.9%)
55+ 6 (1.5%)
Missing 0 (N/A)
Total 407 (100.0%)
171 It can be seen that the large majority of the participants were in the 18 – 24 bracket; this is somewhat contrary to the results of Williams, Yee and Caplan (2008) who found that there were more MMO players in their 30s than their 20s, but the figure in this thesis reflects the large number of student respondents.
Employment Status
Table 5.0.6: Employment Status of the Participants
Employment Status Number of Participants
Student 182 (44.7%)
Employed 142 (34.9%)
Not working 32 (7.9%)
self employed 26 (6.4%)
Away from work / Ill, Maternity Leave 17 (4.2%)
Employer 5 (1.2%)
Government sponsored scheme 3 (0.7%)
Total 407 (100%)
It can be seen from Table 5.6 that the largest group of participants were students, but the demographic data also revealed that the employed also made up a substantial portion of the participants. This indicates that despite the possible heavy time commitments required by MMOs, players will allocate free time to playing such games.
Highest Educational Qualification
Table 5.0.7: Highest Educational Qualification of the Participants Highest Qualification Number of Participants
No qualifications 24 (5.9%)
GSCE’s / O Levels 30 (7.4%)
A Levels 186 (43.7%)
Degree 133 (32.7%)
Masters / PhD 42 (10.3%)
Total 407 (100%)
172 Table 5.7 illustrates that most of the participants had achieved A-levels (or equivalent) as their highest educational qualification. Almost a third had a degree however, with the minority (10.3%) having a higher educational qualification of a Masters or PhD. This finding agrees with that of Williams, Yee and Caplan (2008) also found that the majority of Everquest 2 players were well educated.
Age started playing computer games
Table 5.0.8: Age started playing computer games
Age Number of Participants
Up to 9 202 (50.2%)
10 – 14 128 (31.9%)
15 – 19 44 (10.9%)
20 – 24 11 (2.8%)
25+ 17 (4.2%)
Missing 5 (N/A)
Total 407 (100%)
Table 5.8 indicates that the vast majority of participants (82.1%) started playing computer games before they were fourteen, with a mean age of 10.5 years (SD=6.58), and a minimum of 2 and maximum of 53 years. This would seem to indicate that increasingly, young children are being allowed to play computer games. It should be acknowledged however, that with an average age of 26.56 and the relatively high proportion of student participants (44.7%), a substantial proportion of the sample would have been born with access to tablets and other mobile gaming platforms. These platforms have substantially opened up the gaming genre, so making it much more family, and child, friendly.
173 Primary Server Type
Table 5.0.9: Primary Server Type
Server Type Number of Participants
Normal 158 (38.8%)
Person vs Person (PvP) 126 (31.0%)
Role-play (RP) 79 (19.4%)
Role-Play Person vs Person (RP-PvP) 46 (10.8%)
Total 407 (100%)
The majority of the MMO players played on normal servers (i.e. those where there are no RP rules and no PvP as the default setting), with almost a third playing on PvP servers, as Table 5.9 shows.
Frequencies of Games Played
Table 5.0.10: Frequencies of Games Played
Game Number of Participants
World of Warcraft 307 (75.4%)
Runescape 134 (32.9%)
Guild Wars 124 (30.5%)
Lords of the Rings Online 123 (30.2%)
Rift 100 (24.6%)
Warhammer Online 92 (22.6%)
Eve Online 93 (22.9%)
Aion 80 (19.7%)
Age of Conan 71 (17.4%)
Final Fantasy XI 65 (16.0%)
Everquest 57 (14.0%)
Everquest 2 55 (13.5%)
City of Heroes / City of Villains 52 (12.8%)
Dark Age of Camelot 21 (5.2%)
Runescale 8 (2.0%)
Note: Participants could select multiple games
174 Table 5.10 shows that the most commercially successful MMO – World of Warcraft – was also played by the largest number of participants, with an impressive 75.4% of the sample. It is not the only MMO on the market however and many other games have had some element of the market share. It is also interesting to note that 14.0% of the players were playing Everquest, which was 12 years old at the time of data collection. Given the fast moving nature of the MMO environment, this indicates a strong desire to play a game that was rapidly superseded by other games.
Hours played per Week
Table 5.0.11: Hours played per Week
Hours played per Week Number of Participants
Up to 5 40 (9.8%)
5 – 10 53 (13.0%)
11 – 15 57 (14.0%
16 – 20 71 (17.4%)
21 – 25 58 (14.3%)
26 – 30 43 (10.6%)
31 – 35 28 (6.9%)
36 – 40 20 (4.6%)
40+ 37 (9.1%)
Total 407 (100%)
There is a broad spread of hours played per week, with no overly conspicuous group, although 16 – 20 hours per week was the modal score. However, there are 20.6% of players who play over thirty hours a week, or over four and quarter hours per day on average. Given the number of participants who are students or not working however, this may not be so surprising.
175 Importance of…
In the following section of the questionnaire each participant was asked to rate how motivating the following actions were to them when playing the game. Scores of 1 indicated the action as being of no importance to the player, whilst a score of 10 denoted the action as highly important.
Table 5.0.12: Motivational impact of different elements of game play.
Score Seeing
Note: All scores are percentages, the Modal Score is given in Bold.
From these scores relating to the importance of various aspects of the game, the most obvious point is that there is a sharp discrepancy in motivation between those elements of the game that involve presenting a version of self (Real self and Role Playing), and those that do not. However, it is also evident that Social Interaction is an important motivation for game players, which is a necessary part of the presentation of self, albeit not one that is necessarily consciously acknowledged by a lay audience.
Also interesting is the level high of motivation in completing item sets or unlocking achievements. Dubbed ‘Completism’ in this thesis, it relates to a category of in-game activities which have no relation to high-end game advancement. An example
176 of completing an item set might be getting a particular Helm, Chestpiece and pair of Gauntlets (of a specific set), whilst unlocking an achievement might be getting the Cooking of a particular character up to 450.
With respect to Expressing the ‘Real’ self in the game, the majority of players did not see it as being important in their gaming. It is interesting however that with a mean of 4.29 and with 31.6.% of participants giving it a score of six or more, there is evidence that the expression of self is of some importance to a sizable minority of players. In a similar fashion, the majority of players had no interest in Role Playing though 34.3% showed some interest – a figure that is all the more impressive given that only 19.4% played on a role-play server. Finally, there is a curious split between those players wishing to engage in PvP (Player versus Player) combat, where players fight each other instead of game generated monsters – there is a roughly equal number of players who take absolutely no interest in PvP, and those who are highly motivated by fighting other players..
In order to ascertain if there was any statistical difference between the means for the motivations, a Friedman’s ANOVA was conducted. This was conducted instead of a more usual repeated-measures ANOVA since a Kolmogorov-Smirnov test
indicated that the data was not normally distributed. The results of a Friedman
ANOVA test indicated that there was a statistically significant difference across the six different possible motivations (Seeing high-end game content, Social interaction, Completism, Expressing the ‘Real self’, Role playing, and PvP), X2 (5, n=380) = 432.31, p < .001.
In order to establish how the possible motivations differed to each other, a series of post-hoc Wilcoxon Signed Rank tests were conducted. A Bonferroni
correction was applied and so all effects are reported at a .008 level of significance. The tests indicated a significant difference in motivational impact between some of the factors. Not all will be repeated here due to the repetition this would occur, but a summary is provided, with examples. Seeing End-Game content, Social Interaction and Completism were more motivating for the players than Expressing the Real self, Role playing or PvP (e.g. Seeing End-Game content and Expressing the Real self, z = -12.60, p < .05, r = -.45, or Social interaction and Role playing, z = -13.13, p < .05, r = -.46).
Finally, PvP, though less important than Seeing end-game content, Social Interaction
177 and Completism, was more important as a motivator than either Expressing the Real self or Role playing (e.g. Expressing the Real self and PvP, z = -5.50, p < 0.05, r = -.20).
Results also indicated that there was a non-significant difference in motivational effects between Seeing End-Game content and Social Interaction (z = -.184, p > .05, r = .006), or between Role Playing and Expressing the Real self (z = -.145, p > .05, r = -.005).
Overall, these results strongly indicate that Expressing the Real self and Role Playing were not important to most players, but instead social interaction, seeding end-game content, and completism were significantly stronger motivators.
5.3.2 1st Model - Confirmatory Factor Analysis
Guided by the work of Boomsma (2000) and Jackson, Gillaspy, and Purc-Stephenson (2009), who provide guidelines for the presentation of CFA results, the first CFA is reported here.
Theory
The first CFA sought to examine whether the model developed as a product of the first and second study was an accurate representation of the presentation of self in the underlying gaming population. In doing this, it also sought to answer the research questions of the thesis, as outlined at the start of the chapter. Given the existence of the untested model, CFA was the appropriate type of analysis to use. As with EFA, CFA reduces data down to a number of underlying factors but unlike the former, specifies the structure of those relationships before the analysis. As such then, as Kline (2005) says it is testing a model with stated associations, not conducting an exploratory analysis.
The Set of Models to be investigated
Only one model was created through the previous studies to account for the presentation of self, contrary to the work of Jackson, Gillaspy, and Purc-Stephenson (2009) who suggest that there should be multiple models to compare in the analysis.
Given the grounded theory approach that this thesis used, such an approach would
178 have been seemed contradictory suggesting as it would that the produced model was not the most appropriate fit for the model. However, whilst the use of one model might be unusual for CFA, it does not invalidate the use of this method for testing out the existent model.
Sample
For the first CFA, the full data set was used – that is, 408 participants.
Features of the Observations
The data used in the CFA was continuous, though based on Likert scales with a range from 1 to 5 (1 – Strongly Disagree, 2 – Disagree, 3 – Neutral, 4 – Agree, 5 – Strongly Agree). In this respect, the participant would be asked to say how much they agreed or disagreed with the concept statement drawn from the model.
The Estimation Procedure
The CFA was conducted using Mplus Version 4.0 (Muthén & Muthén, 1998 - 2014), with a complete CFA of the full model being completed. Having done that, recommendations made by Mplus for additional paths and error co-variances were applied – in an attempt to create a model that would fit to the data. In the analysis itself, a Maximum Likelihood estimation procedure was used. The Maximum likelihood procedure is the most widely used method of estimation, followed by Generalised Least Squares (Anderson & Gerbing, 1988).
Evaluation and Model Modification
In the course of first CFA a number of models were tested, although aside from the original primary model drawn from the qualitative study, they were exclusively derived from statistical data. Readers are reminded that the primary thesis model consisted of Elements of self, Social Interaction, Motivations, Nature of MMOs, and Emotionality – in appendix XVI the concept statements, plus their associated CFA names can be found:
179
Model One: Drawn from the qualitative model developed through Studies 1 and 2.
Models Two and Three: Models created as a product of following suggestions made by modification indices. Neither were as successful as Model Four, and so the specific goodness of fit measures are not reported.
Model Four: Final model based on modification indices that successfully ran without serious run errors. As a product of following suggestions made by the Mplus modification indices, the following alterations were made to the model – where ‘with’ indicates a link between concept statements, and ‘by’ a link to a factor:
o Character Creation 1 with Importance of Looks 1 o Gaming History 2 with Gaming History 1 o Character Creation 1 with Spontaneity 1
o Character Infringement 2 with Character Infringement 1 o Identity Re-assertion 2 with Identity Re-assertion 1 o Life Impact 2 with Life Impact 1
o First Character 1 with Character Creation 2 o Emotional Impact 2 with Emotional Impact 1
o Importance of Immersion 2 with Importance Immersion 1 o Moral Consequences 2 with Moral Consequences 1
o Spontaneity 1 with Importance of Looks 1 o Social Referencing 3 with Moral Consequences 1 o Offline Impact 1 with Offline Impact 2
o Replication 1 with Replication 2 o Escape 1 with Escape 2
o Self by Importance of Looks 1
Model Five: A model based on model four, with additional changes made as a product of modification indices. The model failed to converge and was discarded.
Model Six: A Unidimensional model with all indicators loading on to ‘Elements of self’ – this failed to converge and therefore will not be reported.
Model Seven: A model with higher and lower level factors, based on imposed structural model. This is also failed to converge, and was discarded
180 Table 5.13 reports the fit indices for models one and four – the original five factor model, and the five factor model with the outlined Mplus modification indices suggestions.
Table 5.0.13: Fit Indices for CFA Models for the Presentation of self in MMOs
Item Goodness-of-Fit Model One Model Four
Chi square ratio Good < 2
NNFI (TLI) Acceptable > 0.90 Good > 0.95
.591 .724
Note: RMSEA, Root Mean Square of Approximation; CFI, Comparative Fit Index; SRMR, Standardized Root Mean Square residual; TLI (NNFI), Tucker-Lewis Index (Non-normed Fit Index).
The original qualitative model had a poor fit across most of the chosen
measures, aside from RMSEA which indicated an extremely moderate fit. As a product of this analysis, the original model as developed during Studies one and two had to be discarded.
By following the Mplus modification indices, a number of alterations (as indicated above) were made to the structure of the first model, so finally producing model four. These adjustments had the effect of making the model a closer fit to the data, although it is acknowledged that from a theoretical basis, the modifications were not justified. Given that the fifth iteration of the model (again prompted by
modification indices) failed to converge without errors, the fourth model represented
181 the most viable model under the current configuration – but whilst the Chi Square Ratio and SRMR improved so as to indicate an ‘acceptable’ fit, the CFI and TLI were not acceptable. On this basis, and being in mind the lack of theoretical justification for the changes, it was decided to discard the fourth model and to re-analyse the data using an EFA, with a subsequent CFA to assess the results. Given that neither the original model, nor the fourth version achieved an acceptable fit, factor loadings for these models will not be presented.
1st Model Discussion
Overall the first CFA failed to support the qualitative model that intended to provide a coherent account of presentation of self in MMOs. Even with the use of modifications as suggested by Mplus (Muthén & Muthén, 1998 - 2014), the revised model did not meet the appropriate requirements for model fit, so indicating that the original Five Factor model could not be supported. Whilst it would have been possible to have carried on altering the original model based on Mplus suggestions, Kelloway (1998) warns against such a course. His argument is that due to the re-adjustment being post-hoc and empirically derived, that optimally the model should be validated against a new independent sample. Whilst it is acceptable to modify a model and assess the validity on the same data, the interpretation of such modifications is suspect and that parameters added or deleted cannot be said to be confirmed (James and James, in Kelloway, 1998).
Essentially, the risk is that changes made on the basis of a chance variation may not be replicated in a subsequent sample. To alleviate this problem all changes should be theoretically consistent (Kline, 2005), though Steiger (in Kelloway, 1998) warned that researchers can potentially rationalise all beneficial changes to their model, so requiring some self-awareness on the part of the researcher.
So, the original model does not have the necessary accuracy to answer the research questions as established at the start of the thesis and the drawing of
implications is limited. Although the analysis did not provide a favourable fit for the model, which would have been the optimal result, it was entirely necessary and appropriate that a CFA was initially employed. Given the theoretical structure that already existed from the qualitative data, an EFA would have been unsuitable. However,
182 given the lack of fit, the original model must be discarded and an EFA run on the Study three data to establish whether a fit can be achieved.
5.3.3 2nd Model – Exploratory Factor Analysis
Theory
Given the failure of the initial model to achieve a satisfactory goodness-of-fit, even with alterations adopted as suggested by Mplus (Muthén & Muthén, 1998 - 2014), it was necessary to re-analyse the data. Following the work of Kelloway (1998) who argued that an EFA can justifiably follow a CFA model with a poor goodness of fit result, it was decided to conduct an EFA on one half of the data. This would allow a factor structure to be drawn directly from the data, with a CFA being run on the second half of the data to substantiate it. This technique of splitting the sample is entirely valid and found within the literature (e.g. Thompson, Kirk, & Brown, 2006).
Sample
The EFA used the same dataset as that used for the original CFA and as such the participant sampling procedures and action taken on missing data as previously discussed are applicable. Due to the intention to conduct an additional CFA on the results of this EFA, the sample was split into two with the first 204 being used for the EFA, and the 203 being used for a second CFA. Whilst this necessarily reduced the sample size, it is still in line with guidance arguing either for a total number of participants (Kline, 2005; Comney & Lee, 1992, in MacCallum, Widaman, Zhang, &
Hong, 1999) or a variable-to-subject ratio (2:1 in Kline, 2000).
To ensure some degree of randomisation across the split data set, the first participant was allocated to the EFA analysis, the second to the CFA, and so on. This ensured that if any particular participant population were in one part of the database (e.g. the student samples), then they would be randomly distributed across both data sets. Demographics for the EFA sample indicated that they were between 18 and 68 years of age, with a mean of 26.59 (SD=9.180), of which 22.1% were female, and 77.9%
male.
183 Results
Initially, the factorability of the sixty two items was examined. Several well recognised criteria for the ability of a correlation to be factored were used. Firstly, the Kaiser-Meyer-Olkin measure of sampling adequacy was 0.765, above the recommended value of 0.6 (Kaiser, 1974) and Bartlett’s test of sphericity was significant (X2 (1891) = 4873.63, p < 0.01). All but one of the diagonals of the anti-image correlation matrix
Initially, the factorability of the sixty two items was examined. Several well recognised criteria for the ability of a correlation to be factored were used. Firstly, the Kaiser-Meyer-Olkin measure of sampling adequacy was 0.765, above the recommended value of 0.6 (Kaiser, 1974) and Bartlett’s test of sphericity was significant (X2 (1891) = 4873.63, p < 0.01). All but one of the diagonals of the anti-image correlation matrix