Suppose we have to measure how different hierarchy level affect the employee satisfaction. Eg: top, middle & lower
If I have 5 questions regarding employee satisfaction and I want to compare all of them together, which test do you think is most relevant. Do I use regression?
Age: independent variable
Vs
Q1: dependent variable
Q2: dependent variable
Q3: dependent variable
Q4: dependent variable
Q5: dependent variable
You may go far inter item validity test first of all to decide the mutual validity of factors and then for the sake of deciding prominence you may go for inter item Regression Test.
If you are interested in a hierarchical model that maps out the hierarchy (if any) among the five questions, while controlling for the interactive effect of Age Years upon the net response to "All of the questions" (key concept here is "interactive-all"), then use CHAID (chi-square-automatic-interaction-detection) predictive modeling of the sentiment score weights as a summed score.
You did not mention how the Questions were weighted to measure the sentiment of "Satisfaction" but this is what I recommend:
0-Strongly Dissatisfied
1- Dissatisfied
2- Neutral
3-Satisfied
4- Strongly Satisfied
This yields a five point ordinal scale where the summed score for these 5 questions results in a distribution with a minimum value of zero (0) and a maximum value of (20) per survey respondent. CHAID modeling (SPSS) using the summed score as the dependent target variable (DV) and Age, and Q1, Q2, Q3, Q4, Q5 as the independent variables (IV) predicting the DV target, will result in a hierarchical branching tree that will demonstrate how Age and Question interact in explaining the variation in summed score. If you have enough survey responses (n 300 or so) you could try a Binary Logistic Regression (BLR) of the summed score where the distribution has been rendered dichotomous at perhaps the 75th percentile (75th Ntile + = 1, else = 0). The Odds Ratios of the Variables in the Model can be used to establish (and mirror) a similar hierarchy (if any) as that found using CHAID. If you use the same dichotomous (binary) variable in a CHAID model (in SPSS => Analyze=> Classify => Tree), and save the model probabilities (c-statistic) and do the same with the BLR model, you can then plot both model methods in the same Sensitivity-Specificity plot (area under the receiver operating characteristic curve) as a method of internal cross-validation comparison to demonstrate veracity of the findings.
I hope this helps answer your question regarding "how to" discover hierarchy among competing or corresponding topics or themes within a given specific sentiment, opinion, or attitude.
Please read my dissertation available at RG to understand the practical application of one on one and multiple regression analysis. Additionally you shall also be able to understand how to address such cases using qualitative approach as well. The PPT presentation is also available on the dissertation of final defense which shall consume your minimum time.