Tuesday, July 16, 2019

Replicating Costa and Kahn (2011), part II

In my last two posts here and here, I discussed a reproduction and replication of Costa and Kahn's (2011) study on electricity consumption by California households. The difference between the two previous posts is that in one I used data supplied by the authors, and in the other I used data that I downloaded from IPUMS-USA, a publicly-available source. In this post I am reporting new estimates of the same model, but using the most recent data from the 2012-2017 American Community Survey.

This could be thought of as a different type of replication, or perhaps it should be called an extension. The data is from a more recent period, but the models are nearly identical, so you decide. I use six years of data, from 2012-2017, which gives me a similar sample size as in the original study. The results are shown below.

Costa Kahn Extension, Regression Results

Dependent variable:























Adjusted R20.1560.156
Residual Std. Error6.7866.785

Note:*p<0.1; **p<0.05; ***p<0.01

In these regressions, YB1960 (year built 1960) is the omitted category.  The sample includes California households in owned, single-family homes where the householder is between 30-65 years old, as in Costa and Kahn (2011).

One remarkable similarity between this and the results from my previous posts (which included the original study) is the consistency (or robustness) of the coefficient on the price of electricity in the home's "birth decade". Here we see it is -0.239 whereas it was -0.236 ad -0.233 using the author-provided and IPUMS-USA data for year 2000. This suggests that homes built in periods of high electricity periods were built more energy-efficiently.

What do these results suggest for the effectiveness of building codes which were enacted in California in 1978? Note in column one the coefficient on YB1980 is -0.02. This suggests homes built in the 1980s, the period after building codes were enacted in California, use 2% less electricity than homes built in the 1960s. This is consistent with a world where building codes are mildly effective, though of course it doesn't prove it.

Now look in column 2, when the "birth decade" electricity price variable is included. The coefficient on YB1980 is insignificant--homes built in the 1980s after building codes were implemented were no more efficient than those built in the 1960s.

Which of these two interpretations of the effectiveness of building codes is correct? I think the answer to this question comes down to how comfortable one is with the strategy Costa and Kahn (2011) followed in creating the "birth decade" price variable. They used data from only five utility districts and averaged the prices in some cases. Of course this will reduce the variance of electricity prices and make it more difficult to separate any price versus regulation effect. 

So on the whole, despite the robustness of the "birth decade" coefficient estimate across the replications, I tend to think the model in column 1 is more useful as a measure of building code effectiveness. There's not a lot of variation in the birth price variable and even if it is robust across replications, I think this estimate could be picking up the effects of other variables including possibly regulation. I would guess there is a birth price effect and the Costa and Kahn result documents its existence, but I think it is difficult to untangle the precise birth price effect itself here.

At least for my current original research work in progress, where my current focus is on building codes, I plan to estimate models more like column 1 and exclude birth year electricity price effects. I'll estimate some of the models in papers by Levinson, and Jacobsen and Kotchen. The response of home builders to current electricity prices is an important topic give the durability of housing, and future research could improve our understanding of this by exploiting geographically more refined measures of electricity prices. The data challenges here would be formidable.

Stay tuned for more posts related to replication studies using the American Community Survey!

No comments:

Post a Comment