Masters Dissertation Example

3.2.7 Data Analysis 
Responses from the online survey site were downloaded into excel spreadsheets, and matched 
up with the information obtained from the paper by the individual ID numbers. The data was 
checked, and in some cases validated or altered by the ‘further comments’ provided by the 
respondent. Responses of ‘other’ for all questions were checked to see if they could be re-
categorised and if there were any recurring responses. Incomplete responses were deleted if 

---

27 
the respondent did not reach the question that formed the response variable, otherwise they 
were retained.
3.2.7.1 Validation of response sample 
A random sample of 60 papers were selected from the response sample (20 from BC, 10 from 
each other journal) to validate survey responses where possible, such as whether the author 
had made concrete recommendations. A random selection of 100 papers from the full sample 
was taken to validate the characteristics of the papers in the response sample against those of 
the original full sample. The number of citations was recorded for each along with author 
affiliations and residence and whether it was a single species paper.
3.2.7.2 Statistical analysis 
 
Analyses were carried out in the statistical computer program R (R Development Core Team, 
2007). The questions were analysed univariately with the response variable in order to reveal 
any obvious patterns in the data, and chi squared contingency tables were used to test for 
significance between variables. TREE models in R were then used to select the most important 
explanatory variables for multivariate analysis. Data were represented in a series of box plots 
as they give proportional information of the relationship with the response variable (width of 
bar=N for each level of the explanatory variable). 
Due to a mixture of categorical and continuous variables and a binary response variable, data 
were fitted to a general linear model (glm) with binomial errors (Crawley, 2002). For the 
multiple-response questions, each option had to be treated as a separate explanatory variable in 
the analysis. This was not feasible due to the number of explanatory variables, so all of the 
responses for each of these questions were first fitted to a glm and analysed against the 
response variable in order to determine the most important variables for inclusion in the 
model. Similarly, levels within each factor of the multiple choice questions were collapsed if 
the difference between them was non-significant or if they were highly correlated; as indicated 
by a similar slope in the glm. 

---

28 
The variables were tested for both main effects and interactions. The explanatory variables 
identified as most important were included in the model first, and further models were run 
with each different variable to be tested, with terms deleted manually in a step-wise manner if 
an ANOVA test determined non-significance. Any significant main effects or interactions 
were retained in order to obtain the minimum adequate model for explaining the variation 
around the response variable, and hence the most important predictors of the implementation 
of research findings.