Tipultech logo

Data Analysis

Richard Chambers

All data were analysed using the SPSS v.12.0.1 statistical package. Alpha was set at p Ôëñ .05 for significance and p Ôëñ .10 to indicate trends, although exact p-values were reported to allow precise interpretation of results.

The Effects of Gender and Age The impact of gender and age on measures of affect, metacognitive processing, and executive function was evaluated with a series of independent-samples t-tests (for gender) and regression analyses (for age).

Self-Report of Affect and Metacognitive Processing Given the hypothesised ameliorating effect of the intervention on the participants' emotional reactivity, all responses of "3 - I don't get irritated at all by the things that used to irritate me" to item 11 on the BDI provided by the experimental group were treated as anomalies and rescored as "1". Also, participants who reported that they did not drive were instructed to substitute "...drive places on automatic pilot"

(item 12 on the MAAS) with "...walk/ride/take public transport (to) places on automatic pilot". First, between-group differences in baseline (T1) scores on each of the dependent variables were tested. Following this, differences between baseline and outcome (T2) mean scores on the 5 self-report inventories and the DSB were calculated for each group using a series of repeated-measures t-test procedures. Then, in order to establish if the degree of change was statistically significant between the groups, a 2 (group) by 2 (time) ANOVA was conducted for each variable, with the interaction term reflecting the difference in the degree of change between the two groups.

Switching Tasks Data were screened with DMDX's Analyse application prior to statistical analysis. In order to moderate the effect of outliers, RTs faster than 200ms were set to be discarded as outliers, consistent with previous research, based on the assumption that they are outside the typical and acceptable range of processing of word-stimuli (e.g. Kessler, Treiman & Mullennix, 2002). Note that no upper limit was set for rejection of slow RTs, as the DMDX program was set to automatically update words after 15,000ms, in which case the item which was not responded to was summarily scored as an error. No cases were discarded due to this criterion. RTs were windzorised to 2 standard deviations, meaning that any RTs more than 2 SDs from the overall mean RT for each participant were set nominally at 2 SDs. 4.69% of RTs were widzorised overall.

A 2 (task: neutral vs. affective) by 2 (condition: switch vs. nonswitch) by 2 (time) by 2 (group) omnibus ANOVA was initially computed to examine participants' RT measures across tasks.

Measurement of Change There are a number of ways to calculate change scores from pre- and post-intervention scores. The most basic method is to subtract pre-scores from post-scores, to provide difference scores. However, where such change scores are correlated, this is akin to part/whole correlations, which tend to produce misleading results, typically an over-correction of the post-score (Cohen, Cohen, West & Aiken, 2002). Consequently, a linear regression analysis was used here, with the pre-score (T1) used to predict the post-score (T2). This analysis was used to derive a standardised residual score for each dependent variable to represent the change in that variable over the intervention period. These standardised scores were then analysed for correlation with change residuals from other variables in order to evaluate which changes covaried over the intervention period.

Investigation of the Relationship Between Change on Measures of Affect, Metacognitive Processing, and Executive Function Correlations were then calculated to investigate the hypothesised relationships between change in measures of affect, metacognitive processing, and executive function.

Consistent with Hypothesis 4, that the meditation course enhances attention-switching capacity (reflected in reduced switch-trial RTs), only change in RTs for the switch trials were included in these analyses.

Finally, where significant correlations were detected, the relationship between these variables was examined to determine if they fit the following criteria for calculation of mediation tests, proposed by Baron & Kenny (1986): (1) the independent variable (IV) must be significantly correlated with the dependent variable (DV), (2) the mediating variable (MV) must be significantly correlated with both the IV and the DV, and (3) the relationship between the IV and the DV should be significantly reduced when controlling for the MV, using a linear regression analysis. Where such criteria are met, mediation tests were calculated to determine which of the hypothesised mediating relationships between affect, metacognitive processing, and executive control were upheld. Sobel tests were then conducted on the resultant regression coefficients to determine whether each detected mediating relationship was significant (MacKinnon, Warsi, & Dwyer, 1995; Preacher & Leonardelli, 2001; Sobel, 1982). Note that the Goodman (I) statistic was reported, as suggested by Preacher & Leonardelli (2001).

<< >>