Exam 21: Multilevel Linear Models

A data analyst for a software design company was interested in evaluating user satisfaction with a new specialist software for use in medical settings. He designed a multilevel linear model of software users' evaluations within five clinics. What is the contextual variable in his study?

The HR manager conducted her evaluation of staff performance across the ten stores. How many levels of analysis does her study have?

What is hierarchical data?

The data analyst conducted his evaluation of user satisfaction across the five clinics. What is the Level 1 variable in his study?

The data analyst felt more confident and decided to introduce a new parameter, Hotel (the 8 hotels in the chain). This parameter is what sort of variable within the model?

The data analyst decided to run a five parameter multi-level linear model, using the same outcome variable Satisfaction, which measured guest satisfaction on a ten-point scale. He now had five parameters in his model: Satisfactionx (guest' satisfaction with the hotel chain prior to them staying, measured on a ten-point scale); Sex (guests' gender); Age (guests' ages); Hotel (the 8 hotels in the chain); and Duration (measuring duration of each guest's stay in days). The log-likelihood for the five-parameter model was 1689. How does this compare to his previous model?

As a business development manager for a car rental company, you are attempting to conduct a multilevel model of performance of sales people within and across four car showrooms in two different regions, but you keep facing the problem of multicollinearity. How might you solve this?

The HR manager of a chain of department stores wants to evaluate staff performance across the chain. There are ten stores in the chain and each store has six departments (Household, Food, Technology, Women's Clothes, Men's Clothes and Children's Clothes). She is interested in the performance of staff within and across departments and stores in the chain. What sort of analysis would be most appropriate?

What assumptions apply to multilevel linear models?

What are growth models?

How can we overcome the problem identified in the previous question?

What assumption of linear models does multilevel linear modelling violate?

A data analyst for a small chain of boutique hotels was interested in what shaped guest satisfaction with their hotel stays. He decided to build a multi-level linear model. Because he had not run one before he decided to build the model in stages, starting with a basic linear model. His outcome variable was Satisfaction, which measured guest satisfaction on a ten-point scale. His initial parameters were Satisfactionx (guests' satisfaction with the hotel chain prior to them staying, measured on a ten-point scale) and Duration (measuring duration of each guest's stay in days). He ran the model and found that Satisfaction2 had a p = 0.05 (b = -1.69) and Duration had a p = 0.01 (b = 0.665). How would you interpret this?

The data analyst found that the introduction of 'Hotel' to his model altered the log-likelihood from 1798 (the previous model) to 1786. How would you interpret this?

The data analyst wanted to finalise his model for guest satisfaction with the introduction of one last parameter, Reason (a dummy variable of guest reason for staying at the hotel, coded 0 for holiday stay and 1 for business stay). How might we categorise this parameter within this model?