What Factors Affect Office Temperature? Creating Custom Maps
This post is the third in a series that shows how we conducted an observational study about office temperatures, from design to execution to analysis and visualization. See Part I and Part II.
In the previous two posts, we discussed the design and execution of an observational study about what factors affect office temperature. In Part II, you saw how we actually collected the data and how we cleaned the data before analyzing it in JMP. Today we will showcase some new features in JMP 9 to visualize some of our results, and we will run a model that will help us determine what factors are strong predictors of office temperature.
One of the new features in JMP 9 is the ability to create custom maps. We created a map of the 4th floor of Building S here at SAS, where most of the JMP staff is located. Using new features in Graph Builder, we can analyze the office temperature results visually.
The map shown below is the JMP floor, grouped by time of day, with color reflecting the Fahrenheit value. This report also contains a background image (another new feature in JMP 9) that happens to be the blueprint for the 4th floor of the building. Exploring data visually in this way can give us hints as to what factors are affecting office temperature. Looking at this particular map, it looks like the offices on the east side of the building are warmer in the mornings than they are in the afternoons. On the western side of the building, the opposite appears to be true. From this visualization, we might expect that both of these variables are affecting office temperatures, or perhaps that the interaction between these terms is significant.
Next, let’s take a look at Fit Model to see the numerical analysis results. I entered all the main effects into the model for the first try. In the table below for Fixed Effect Tests, we see a number of variables that may have a significant effect on temperature, including thermometer, outside temperature, type of space, sector, volunteer and outside conditions. The most surprising thing to me is the fact that neither east or west nor time of day appears to be significant. This first version of our model does not measure interactions between main effects, but my next step will be to include an interaction term in the model for east or west ∗ time of day, since my Graph Builder visualization above indicated that there may be an effect.
The next table of Fixed Effects Tests shows a later version of my model, where I have removed wing and added the interaction term east or west ∗ time of day. Here, my suspicions are confirmed. The interaction term is clearly significant and even has the highest F Ratio.
We want to eliminate the insignificant factors from the model, and the method I chose is a Stepwise regression. The following factors wind up being significant when Stepwise is finished. Notice what is missing: The outside weather conditions and temperature do not seem to have any effect on the inside office temperature. Note that the interaction term is still showing as significant -- this is an example of how visualizations can sometimes guide you in making decisions during your analysis.
In the next post, look for more visualizations of this data using different JMP tools.
How did you like the new features in Graph Builder? What other visualizations would you be interested in seeing?