Tuesday, April 4, 2017

A conclusive non-conclusion about dimming in the AAVSO data

I'm spending too much time on this, so will have to bring it to a close until the summer's observing is done.

I took one more look at the AAVSO data, this time doing something called binning, similar to what Brad Schaefer did with the DASCH data in his paper on dimming in the historic photographic plates. Binning takes several observations within a defined time period and averages them before attempting to fit a model to them. In this case, the model is a simple straight line. This has the effect of giving each time period an equal "vote" in the best fit to the model, even if there is much less data in one time period than another. In the case of the AAVSO data, some observers would report many observations over a short period of time, which tended to overweight their observations. Binning mitigates that.

Of course, you have to decide what period of time you will use for binning the roughly 500 day span we have so far. I arbitrarily picked 10 days, and averaged the observations for each observer over that time period. There were 47 AAVSO observers in all whose measurements survived the filtering process in the "V" passband.  There were 48 observers, but I identified one who temporarily had apparent problems with respect to the others, so was filtered out to make it simple.
The V Band Fit with 10 day binning

While there is a dimming trend you can pick out of the binned data, the statistical significance is unimpressive. It's not a slam dunk. The dimming shown here is about 0.68 magnitudes per century, or about 0.6% per year. It could be real, but as you can see, it's small with respect to the scatter in the data. Here's what you get when you go to 50 day bins with the same 47 observers:
The V data with 50 day binning
The result is roughly the same, but the dimming is a bit higher: about 0.9 magnitudes per century. The other bands I looked at (I,R, and B) have fewer observations, and binning made the results statistically insignificant.

I will content myself to wait another 200 days or so to see if this is consistent. There are a couple of reason for this:

  1. There is plenty of reason to suspect that a linear dimming is too simple. It could stop for a while, or even reverse. We just don't know what is going on in any detail, and there is no physics in a linear fit. It's still possible that no dimming has been going on lately.
  2. We might be fooled by uneven noise and biases, or what is called "systematics" in the data. Data over a longer time span can help us sort these out.
So, stay tuned. If you have your own analysis, come on over to Reddit and share it with us.