Since energy efficiency is one of our primary goals at Sense, we are always looking for new ways to understand the factors that promote such efficiency. And conversely, we’d like to understand some of the drivers behind high energy consumption. An article from Greentech Media on the differences between energy consumption across the fifty states caught our attention. Since there are many variables that differ systematically across states, this seemed like an interesting way to gain relevant insight on the drivers of energy efficiency. That article reported that the political preferences of the states (as measured by the CO2 Scorecard) are an important variable in how much energy citizens consume on average. This got us thinking about the most important factors that are likely to influence differences in electricity consumption across the fifty states.
Given that the list of influences on electricity consumption is quite long; we decided to focus initially on the following five factors:
We used linear regression to model the relationship between the average monthly electricity consumption by state and the five factors listed above. First, we studied the effect of each factor independently, and we quantified the “goodness” of the resulting linear model using a metric known as the coefficient of determination1. Next, we combined various groups of variables and looked at how well the combined sets predicted consumption.
The first set of results is summarized in the graphic above/below. We see that the price of electricity is the most important single factor in energy consumption, with a coefficient of determination (R2) of 0.536 in our data sets. Political preferences2 (“red vs blue states”) and summer temperatures are the next most important factors, with measured R2 of 0.389 and 0.381, respectively. Winter temperatures are significantly less important, yielding an R2 of 0.274, and average household income was the weakest factor (R2 = 0.151). For the data geeks, all of these correlations were statistically significant using a two-tailed F-test with 0.05 significance level.
Of course, our various input variables are not mutually independent, and in the second step we found that “price level” and “red vs blue” variables explain much of the same variance. That is, the R2 value increases only slightly (from 0.54 to 0.58) if the voting variable is used in conjunction with the price variable. Summer temperatures, in contrast, are an important independent variable; when they are combined with the pricing variable, the coefficient of determination increases to 0.71. Beyond these two variables we gain little from other factors in our set, as linear regression with all five variables yields R2 = 0.76. In other words, whether a state uses a high amount of electricity is primarily determined by the price of electricity and high summer temperatures. The results are summarized in Fig. X below.
You may remember your stats’ prof’s mantra “correlation does not imply causation!” That said, our observations suggest that although political preference may appear to be a driving factor in explaining variations in electricity consumption, price and summer temperatures are in fact the primary determinants. The effects are disguised by the fact that blue states tend to have higher electricity prices, a good topic for a follow-up article. It would also be interesting to analyze how electricity prices and summer temperatures have affected energy consumption over time.
So, as you take a look at weather forecasts this summer and head out on your 4th of July weekend, take a look at your Sense usage and tell us: what are the primary factors affecting your electricity bill? Happy 4th of July and happy Sensing!
1 The coefficient of determination is a number that indicates how much of the energy consumption, is explained by each independent factor, e.g. the price of electricity. The higher the coefficient of determination, the better.
2 Political standing is scaled. So, instead of just listing red and blue states, we scaled the data based on electoral college stats. This shows how “blue” a state is. Hawaii, a really blue state, scaled at 0.7; whereas, Louisiana, a very red state, scaled at 0.4.