• No se han encontrado resultados

Encuesta dirigida a Instructores

5. Descripción del proceso de formación del Servicio Nacional de Aprendizaje SENA en el

6.4 Encuesta dirigida a Instructores

The design of the study was cross-sectional (between subjects). The data were checked for assumptions prior to analysis to determine the most suitable methods to employ in order to test the hypotheses. The main assumptions of 1) dependent variables measured at the interval or ratio level, 2) independent variables consist of two or more categorical, independent groups, 3) independence of observations, 4) adequate sample size, 5) no univariate or multivariate outliers, 6) multivariate normality, 7) linear relationships between pairs of dependent variables, 8) homogeneity of variance and 9) no multicollinearity are outlined below with further details.

1) The dependent variables should be measured at the interval or ratio level.

The Portrait Values Questionnaire (PVQ), Vancouver Obsessive Compulsive Inventory (VOCI), Symmetry, Ordering and Arranging Questionnaire (SOAQ) and Responsibility Attitudes Scale (RAS) are all interval based measures with responses given on a scale. Checking behaviour was measured by time taken to complete a letter identification task (in seconds) and so is interval in nature and the ordering measure is based on the number of pens placed in certain positions with scores falling from 0-12. Cleaning behaviour is ratio (number of participants who do use a wipe versus those who do not).

2) The independent variable should consist of two or more categorical, independent groups.

All independent variables consisted of two or more categories from independent groups of participants. There were different independent variables investigated depending upon the hypotheses being tested. For example Hypothesis 1 focused on the independent variable of the three groups (conservation, openness and control)

112 whereas Hypothesis 3 focused on high obsessionality participants versus low obsessionality participants.

3) There should be independence of observations

Participants were randomly allocated to the main experimental groups (conservation

prime, openness prime and controls) which were mutually exclusive. All other

groups used i.e., high versus low obsessionality were also mutually exclusive. As such, there were different participants between each group and there was independence between the observations within each group.

4) There should be an adequate sample size

The power analysis previously completed (see method section 2.3) indicated that there was sufficient power. The power analysis estimated that 86 participants were necessary in order to show sufficient power based on previous similar experiments (e.g. Maio et al, 2009). The current study had 89 participants (29 in conservation group, 30 in openness group and 30 in the control group) meaning that there was a sufficient sample size.

5) There are no univariate or multivariate outliers

This is previously outlined and was dealt with in section 3.2.4 (above)

6) There is multivariate normality

Skew and kurtosis scores were obtained for all of the measures (aside from the PVQ and the cleaning behaviour, for the reasons outlined above) as were Q-Q Plots and histograms to study the distribution of the scores (see Appendix 21-26). Kolmogorov-Smirnov tests were also carried out for each measure and were interpreted in conjunction with the other information available (as recommended by Field, 2013). This was initially done for the measures as overall scores (using the scores from across the three experimental groups) before being repeated for scores

113 within the three experimental groups separately. The interpretation of this analysis for each measure is outlined below.

Vancouver Obsessive Compulsive Inventory (VOCI)

A Kolmogorov-Smirnov test indicated that the distribution of the overall VOCI scores deviated significantly from normal (D(89) = .149, p<.001) and this was supported by studying the relevant Q-Q plot as well as skew and kurtosis scores (see Appendix 21). There was particularly significant positive skew (z = 3.557, p<.001) within the overall measure. There were mixed results when the VOCI was investigated using the individual groups with the conservation group (D(29) = .208, p<.01) and the

control group (D(30) = .210, p<.01) both showing significant deviation from normal.

The openness group (D(30) = .148, p=.09) was found to not deviate from a normal distribution. However, within the openness and control groups the distributions of scores were found to be significantly positively skewed (p<.05 and p<.01 respectively) suggesting that normal distribution should not be assumed.

Symmetry Ordering and Arranging Questionnaire (SOAQ)

Indications of significant positive skew were seen in the overall measure (z = 2.894 p<.01) and within the conservation and openness groups (z = 2.002, p<.05 and z = 1.991, p<.005 respectively). Studying the Q-Q Plots and histograms suggested that the distribution could be close to normal in some of the subgroups (see Appendix 22). Kolmogorov-Smirnov analysis indicated that the distribution of the overall SOAQ scores deviated from normal (D(89) = .121, p<.01). However, the individual group scores suggested that the SOAQ did not deviate from normal (conservation, D(29) = .177, p=.02; openness, D (30) = .148, p=.09; control, D(30) = .130, p=.20). This discrepancy could be due to the sample sizes as the K-S test has been shown to be overly stringent with larger sample sizes where very small deviations from a normal distribution can lead to significant results. However the significant skew scores for the conservation and openness groups suggest that normal distribution should not be assumed (Field, 2013).

Responsibility Appraisal Survey (RAS)

The RAS in contrast had very small positive skew scores and small negative kurtosis scores generally across the three groups indicating that this scale was closer to

114 being normally distributed than the VOCI and SOAQ and this was supported by examining the Q-Q plots (see Appendix 23). A subsequent Kolmogorov-Smirnov analysis (D(89) = .086, p=.10) indicated that the distribution of the overall RAS scores did not deviate from normal. The distribution across the individual group scores were also found to not deviate from normal (conservation, D(29) = .083, p=.20; openness, D(30) = .112, p=.20; control, D(30) = .145, p=.10). Therefore the RAS can be considered to have a normal distribution.

Checking behaviour

There were general positive skew and negative kurtosis scores within the checking behaviour measure which approached significance levels although the histogram and Q-Q Plots suggested that the overall measure was close to normally distributed (see Appendix 24). Kolmogorov-Smirnov analyses indicated that the overall checking measure did not differ significantly from a normal distribution (D(89) = .091, p=.067). Within all three of the groups the Checking behaviour measure was found to not differ significantly from a normal distribution (conservation, D(29) = .072, p=.20; openness, D(30) = .150, p = .09; control, D(30) = .148, p<.10). Considering all of the information available a normal distribution could be assumed for this measure.

Ordering behaviour

The Q-Q plot along with the significant positive skew score (z = 2.447, p<.05) and a negative kurtosis score which approached significance indicated that there was significant deviation from a normal distribution across this measure (see Appendix 25). Supporting this, Kolmogorov-Smirnov analyses suggested that the distribution of the overall ordering behaviour measure differed significantly from normal (D(88) = .201, p<.001). This was also the case within the conservation group (D(29) = .172, p< .05), the openness group (D(30) = .271, p<.001) and the control group (D(30) = .160, p<.05) which all showed positive skew and negative kurtosis although only the openness group skew score reached significance (p<.05). With this in mind a normal distribution could not be assumed.

115 7) There is a linear relationship between each pair of dependent variables for

each group of the independent variable

Scatterplot matrices were obtained for all of the variables comparing the questionnaire measures against the behaviour measures at the group level (i.e. for the conservation, openness and control groups). Indications were that linearity could not be assumed (see Appendix 27) with r2 scores of 0.058 at the highest and 0.003 at the lowest. Linearity refers to whether the amount/rate of change, between scores on two variables is constant for the entire range of scores. If the relationship between the variables is non-linear then using statistical methods that assume a linear relationship will underestimate the strength of the relationship, or will fail to detect the existence of a relationship (Field, 2013).

8) Homogeneity of variance

The homogeneity of variance was also assessed in the questionnaire data. Levene’s test showed that for the VOCI and the SOAQ homogeneity of variance could be assumed (VOCI, F(2, 86) = .691, p = .50; SOAQ, F(2,86) = .531, p = .60). Whereas for the RAS the Levene’s test indicated that the variances of scores were significantly different between the conservation, openness and control groups (F(2,86) = 6.003, p<.01) meaning that for this measure homogeneity of variance could not be assumed. The checking and ordering behaviour measures both showed that homogeneity of variance could be assumed (Checking, F(2,86) = .909, p = .41; Ordering, F(2,86) = .057, p = .95). When the variance ratios were calculated it was shown that none of the ratios for any of the measures fell above the critical value (F(2,86) = 3.09) suggesting that the Levene’s tests might be overly stringent (Field, 2013). Generally homogeneity of variance was therefore assumed.

9) There is no multicollinearity

The correlation matrix below in Table 3.6 (section 3.4.2) shows that the correlations between the questionnaire measures are between -0.496 & 0.586 which are significant correlations. This indicates that there could be difficulties with multicollinearity if regression analyses are used as two closely related independent

116 variables could cause the predictive value of the regression model to be inflated. However, there are simple strategies for dealing with these difficulties and Field (2013) recommends removing one of the strongly correlated variables from the analysis to ensure that the regression model is not biased.