David W. Allen

I recently received my Nest Learning Thermostat after they were sold out of the initial manufacturing run. There has been plenty of articles written about how nice the design is, how easy it is to install, and its Apple pedigree. While those are all great (it is beautiful, it took 10 minutes to install, and I’m an Apple Fanboy) I was really interested in the sensors.

Specifically, I had a suspicion that the Honeywell thermostat thoughtfully mounted on an outside wall next to a window was really reading the wall temperature and not the ambient room temperature. The Nest seemed to have an array of sensors which could work together to more accurately differentiate the room temperature from it’s own internal temperature.

Before installing the Nest, I took 10 temperature samples over the course of an hour (n=10) of the room temperature and the wall temperature around the thermostat. I also recorded the stated room temperature of the thermostat. All measurements were taken while the heater was not running, but these are by no means “accurate”.

After installing the Nest I repeated the temperature readings. There are some outliers in the data which I attribute mostly to the measurement equipment or perhaps taking a reading directly after a heat cycle. I am reporting here the min, max, and median DIFFERENCE from the thermostat’s stated room temperature and the readings taken by me.

For the Honeywell:

• Wall Temperature: [-0.7, 0.5], median = -0.3
• Room Temperature: [-5.2, -3.9], median = -4.8

For the Nest

• Wall Temperature: [-3, 3.9], median = 2.8
• Room Temperature: [-3.0, 0.8], median = -0.8

*Readings are in °F

So it seems that qualitatively I was correct in thinking that the Honeywell was mostly measuring the wall temperature while drastically under reading the room temperature. I had typically set the Honeywell for 64° feeling that it ran about 4° warmer.

The Nest had some outliers in the measurements (taken directly after install) but the later data readings settled down. This normalization can be seen by the median of the measurements. The Nest seems to do an excellent job of reading the actual room temperature and ignoring the mounting surface as I had hoped.

Human perception is the least reliable measurement ever, but I do feel there are less drastic temperature swings in the house now. I attributed the swings before to things like the outside wall getting direct sunshine, being in the shadow, wind, etc.

So for an initial impression, the Nest seems to be doing what I bought it for… accurately controlling the temperature in the room. I’ll write again after I’ve had time to determine if the “schedule learning” is awesome or if we just end up futzing with it a lot. Our family has an odd/flexible schedule.