Chapter 7 Cross-sectional Study: Current members
7.4.1 Preliminary data screening
Various steps were taken before the inferential analyses were conducted; ensuring data accuracy, dealing appropriately with missing data, reliability estimation and computing ‘true’ mean scores.
Data accuracy
According to Tabachnick & Fidell (2007), inspection of univariate descriptive statistics for accuracy of input; out-of-range values, plausible means and standard deviations should be the first steps taken in data screening. When completing the questionnaire, all questionnaires were entered using Survey Monkey, even the paper-based ones. As such, there was no option for numbers to be freely entered into, but instead could only be selected from a range of scale options. Although more time consuming, this ensured that data could not have been manually entered out of the given one to seven range, as is the risk when entering paper-based questionnaires directly into a spreadsheet whereby the operator may be at risk of entering any number.
Missing data
Tabachnick & Fidell (2007) also state that missing data must be dealt with before any inferential analyses are conducted. In essence there are two considerations that should be made regarding missing data; what data (if any) should be recoded as missing, and what the implications are for genuinely missing data. Regarding what should be recoded as missing data, as mentioned above, any scores that were missing, or not easily identifiable from the paper-based questionnaires received were re-coded as missing data. The perceived service quality items were the only items which had a ‘not applicable’ option. Any ‘not applicable’ responses to the perceived service quality items were treated as missing data.
Reliability estimation
Because mean scores were to be computed of each predictor, it was essential assess the internal consistency of the items representing their associated predictor. The Cronbach alpha values of all scales were obtained in order to test internal consistency of the items scores representative of a given predictor.
Although the ideal Cronbach alpha value of a scale should be 0.7 or above (Kline, 1999; Pallant, 2005), a lower Cronbach alpha value of 0.6 is considered an acceptable threshold within the domain of social science, especially for research purposes (Cohen, 1988). Most scales were internally consistent to an alpha level of 0.6. Further, most alpha levels were in excess of the minimum 0.6 threshold (Table 7.3). As such, mean scores of the items could be generated with a certain degree of confidence.
Computing of ‘true’ mean scores
To ensure even further confidence in the mean scores, all of the mean scores for the predictors were only computed if all items in the scale had been responded to. This is different from other ways of computing the mean score, which may allow the mean score to be generated from however many items in the scale had been responded to, even if not all items had been responded to. This is risky, as it allows a mean score to be computed which, potentially, may not be based on the full number of items deemed necessary to represent that particular predictor, therefore not appropriately measuring the construct.
Whilst this increased the validity of the mean scores, this inevitably reduced the sample available for inclusion into the analyses; if a participant’s mean score was not computed for a particular predictor, then that participant would not be included in an analysis of that particular predictor.
The mean scores and standard deviations are displayed in Table 7.3 (correlations between the variables can be found in Appendix C3).
Table 7.3 Descriptive statistics
Cancelled Retained
α N Mean SD N Mean SD Intention to cancel .85 68 2.51 1.93 574 1.60 1.07 Positive word-of-mouth .93 69 4.51 1.70 569 4.68 1.47 Perceived service quality - Staff .89 27 5.21 1.13 280 5.31 0.99 Perceived service quality - Classes .87 54 5.45 1.04 426 5.46 0.96 Perceived service quality - Changing Rooms .75 79 5.64 1.11 619 5.59 1.02 Perceived service quality - Physical Facilities .83 68 5.57 0.98 539 5.62 0.81 Perceived service quality - Gym Environment .88 73 5.68 1.12 564 5.84 0.83 Perceived service quality - Pool/Spa .72 74 5.56 1.11 561 5.67 0.95 Perceived service quality - Bar/Café .78 67 5.58 1.00 512 5.54 1.07 Perceived service quality - Childcare .89 25 5.64 1.09 156 5.58 1.16 Perceived value for money n/a 71 4.51 1.49 584 4.77 1.50 Brand attractiveness .73 68 4.85 1.24 574 4.86 1.23 Brand prestige .76 68 4.57 1.46 575 4.26 1.49 Brand distinctiveness .76 68 4.96 1.30 574 4.85 1.25 Brand similarity .74 71 3.77 1.47 583 3.75 1.51 Individual stereotyping .71 68 3.52 1.17 574 3.45 1.22 In-group homogeneity .61 71 3.76 1.23 584 3.67 1.26 Rapport - Staff .90 68 6.36 2.72 573 6.79 3.09 Rapport - Members .90 68 3.63 1.42 572 3.94 1.65
State anxiety - Staff .80 68 2.60 1.34 574 2.05 1.21 State anxiety - Members .82 68 2.70 1.33 573 2.18 1.32 Social physique anxiety .84 68 3.83 1.28 572 3.42 1.19
External regulation .76 68 1.70 0.71 572 1.53 0.78
Identified regulation .67 68 5.37 1.07 574 5.39 1.04 Integrated regulation .77 68 3.98 1.41 574 4.24 1.32 Intrinsic regulation .84 68 5.46 1.18 569 5.71 1.00 Introjected regulation .77 68 4.23 1.27 571 3.94 1.59
Further with regard to the intention to use predictor, it was essential to control for any barriers that will affect, or have affected, the ability for members to use the club in their usual capacity. For instance, it was essential to try and control for fitness club members’ absenteeism from the club due to holidays etc. Given the time of year that the research was conducted (July-August), there was perhaps more likelihood that fitness club members may have been on, or were due to go on, holiday. Therefore, just asking members how often they intended to use the club or had previously used the club would have been misleading. For example, a member may have answered ‘4-6 times’ which over a 4 week period would suggest a minimum usage of once per week. However, if they had planned to be absent from the club due to a two week holiday then intending to use the club 4-6 times in just two weeks over the month suggests a minimum usage of twice per week.
This was dealt with by allowing the responses of members who had indicated foreseeable absence to not have their intention to use the club score computed, instead being recoded as missing data.
After the preliminary data screening steps had been taken, and the mean scores appropriately generated, the inferential statistical analyses were conducted.