I am designing a research in the area of enterprise risk management (ERM) and its impact on firm profitability in the insurance industry. There are 30 companies in the industry. My plan is to test how managers’ (300 managers who engaged in ERM) perception on level of ERM practices impact on profitability of their companies. Managers perception on level of ERM practices (independent variable) will be measured through a questionnaire. I would like to obtain actual profitability figures (dependent variable) from the company’s annual reports. Multiple regression is proposed as the method of analysis.
I have a doubt about using primary data collected from a questionnaire (ERM practices, obtained from 300 managers) and secondary data (real profitability figures, obtained from annual reports of 30 companies) to test hypotheses. Same profitability figures will be repeatedly used against the respondents of the same company (Example, if 10 respondents from company ‘A’, same profit figures of the company ‘A’ will be used against all the respondents of the company).
I seek experts' advice to resolve the above methodological issue. Thank you.
While firm profitability from the audited financial statements will be reasonably objective, subject to accounting estimates (eg provision of future estimated liabilities), the questionnaire might be less objective. Therefore I suggest you think hard about questionnaire design as regards the precision of language and scoring, as this will need justification in your research. If you are using Likert scaling I would avoid making the range of scores too large or it will be difficult to distinguish different responses accurately. Might be more objective to list ERM features and tick yes/no/NA to keep the scoring as objective as possible.
Fundamentally, I would recommend a more precise definition of "perception" - itself a rather subjective term! My experience in observing poor experimental analysis is that in many cases the problem is trying to do TOO MUCH with a single hypothesis statement. There is often a confusion between the strictly binary requirement of a statistical hypothesis and the general usage of "hypothesis" as any conjectural statement.
You can always add additional hypotheses to clarify an issue as you go along. A good hypothesis in statistical analysis should have a very clear "either/or" statement that can be analyzed with the collected data. FIRST define clearly what you mean by "perception" in terms of objective factors (that may take more than one hypothesis itself), then similarly define your "success" criteria. Profitability may be one measure, but not necessarily the metric used by all of the firms. That assumption must be verified. Many companies use different metrics for success that may affect how their reports show up in your surveys.
- Badges
- Science topic