A strength of this study is that it is the first to specifically consider the utility of a specific attribution theory in relation to the general label of personality disorder. The comparison between the diagnostic groups of personality disorder and
schizophrenia on the different aspects of this pathway model of stigma is also novel.
Schizophrenia was chosen as the control variable because, in the existing literature, it is one of the two most widely used diagnoses for comparison with personality disorder (the other being depression). This is because, individuals with these diagnoses are also often thought of as ‘difficult’ patients (Koekkoek, et al., 2006). Schizophrenia was chosen as it is considered a more severe and enduring mental health diagnosis than depression. It was, therefore, thought that the inclusion of schizophrenia as the control diagnosis would allow for greater ease of comparison with previous research findings.
However, it should be considered that there are important differences between the two diagnoses, not least, that schizophrenia has a well researched, strong evidence base for effective treatment (NICE, 2009). This contrasts with personality disorder for which the evidence base for effective treatment has only recently begun to emerge.
Another important difference is that medication is often seen as a frontline treatment for schizophrenia (NICE, 2009), whereas it is used much less frequently to manage symptoms of personality disorder. Although these differences did not appear to have a significant effect on the attributions, emotions and intended behaviours of student mental health nurses, it may be that the effect would be more marked in a sample of qualified nursing staff who are likely to have more knowledge and experience of the causes and treatments for each of these diagnoses.
4.4.2.4 Sample.
The use of student mental health nurses in this research may be considered a strength. This is because there are very few studies that consider the attitudes and attributions made by student mental health nurses. By expanding the literature in relation to this participant group, it allows for comparisons to be drawn with previous studies which have used qualified staff, which will enable consideration to be given as to how these attitudes are mediated by the factors of exposure, experience, knowledge and interest. However, a limitation of the study is that a robust measure of participant experience was not used, therefore meaning that it is not possible to accurately assess the effect of experience on attributions and intended behaviours. This is a significant limitation of the study as familiarity and experience are cited by Corrigan et al. (2003) as an influencing factor. Whilst a basic measure of experience (number of years into training) was completed and Chi-square analysis did not reveal any significant difference in experience between the groups, this measure is not robust enough to determine whether there were any actual differences in experience between the groups. It also does not take into account the type of experiences had. This is
important as learning theories would suggest that past experiences would be likely to influence the development of attributions and future decisions about intended
behaviour. Future studies might consider using a more in-depth measure of experience in order to address this limitation.
4.4.3 Analysis.
It is also helpful to consider the strengths and weaknesses of this study in terms of the statistical analysis of the data. As previously reported, several of the variables were severely skewed. This meant that, for hypotheses one to three, multiple ANOVAs were conducted in addition to Pearson’s chi-squares and Mann-Whitney U analyses, whereas if the data had been normally distributed, multiple analyses of variance (MANOVAs) could have potentially been utilised, although interpretations of these findings would have been much more complex. The increased number of individual analyses therefore increased the chances of Type I error (i.e. concluding that there is statistically significant effect where one does not exist). In order to address this issue, follow up t-tests were conducted for each significant statistical finding. Whilst it is acknowledged that other post-hoc testing approaches could have been used, for example Bonferroni approaches, it is considered that the small number of t-tests conducted would not have had an effect on multiplicity. This limitation with regards to Type I error should also be held in mind when considering the results for hypothesis four.
This study also used a large number of Spearman’s rho correlations, once again increasing Type I error, whereby the presence of a significant correlation is erroneously concluded. Again, whilst Bonferroni approaches could have been used in order to correct for multiplicity, this would result in the alpha levels becoming very small (i.e. Field (2012) suggest that for a significant correlation to be found it would
require an alpha level of .0001). This would then be likely to result in a Type II error where there is a failure to reject a false null hypothesis.
Another limitation arising from the severely skewed nature of the data is the fact that, in order to conduct any form of meaningful analysis on the items that did not meet the assumptions of normality, dichotomisation was required. It is assumed that this is also the reason that data were dichotomised in the study conducted by Purves and Sands (2009). This dichotomisation results in two main issues: firstly, the fact that information is lost during the dichotomisation process; secondly, the fact that participants were unaware that scores would be dichotomised in this way also presents a difficulty. For example, the data for the anger variable was dichotomised in such a way that scores of four and below represented a ‘low’ score, and scores of four and above represented a ‘high ‘ score. However, due to the semantic differential nature of the scale, each increment on the scale is subjective and so, for example, someone endorsing a score of five on an item may not have perceived this to be a high score. Thus, in the dichotomisation process, the meaning of individual scores may have been altered. The fact that this cut-off has had to be implemented may also not be ideal for use with a semantic-differential scale, as it becomes unclear what is truly meant by low and high scores at the time of completion of the questionnaire. Also, it is likely that different results would have been found if this cut-off was placed elsewhere, for example if the anger variable had been dichotomised at five as opposed to four. This should be considered as a potential concern for future studies in this area where semantic differential scales are being considered.
In terms of additional analysis, it may have been interesting to repeat the analysis in relation to experience to determine whether attributions, emotions and intended behaviour change over the course of training. This would have allowed a
comparison with a study by Markham (2003), where it appeared that individuals tended to become more negative about service users with personality disorder over time. Whilst it would have been possible to do this analysis with the current sample, it is unlikely that adequate power would have been achieved. This is something that could be addressed in future research.