A few months ago I wrote a blog post about Rupert Sheldrake about his ideas on the extended mind. I honestly found his ideas pretty silly, but his experimental design really caught my attention. There was one experiment he talked about where they observed a dog anticipating the owner returning home. My first few thoughts about how this experiment would produce false positives seemed to be taken care of by how he designed the experiment, for example, I thought it was probably just the case that the owner comes home at the same time every day, but they did an experiment with the owner returning at a random time. I also thought that perhaps the dog just gets up and waits at the door every five minutes, but this was supposedly not the case. I was intrigued and said that I would be interested in looking at the experiment in a little more depth, and that is what I want to do today.
You can see the paper itself here. I'd also like to say up front that I am not a scientist (I suppose I consider myself a science fan) so take anything I say here with a grain of salt. And if you see anything I have said that you think is wrong please feel free to speak up, discussion is always welcome.
The paper begins by claiming that it is a common event that people notice their dogs anticipating their owners return, for example they might wait by the door. Apparently it is common for people to attribute some kind of a psychic link between the dog and owner to explain the behavior. This paper seeks to test that hypothesis against 4 conventional explanations
1. The dog could smell or hear the owner approaching
2. The dog could be responding to the owner's routine
3. The dog could be reacting to subtle cues from the people who are home
4. The dog could be reacting that way all day and people have selected memory
To account for these conventional explanations, the experiment will
1. Not include the last 3 minutes of the trip in the data
2. Randomize return times
3. Not allow the people at home to know the return time
4. Standardize the tracking of the dog and have people record blinded.
This all seems good so far.
The owner left the house for a while, not knowing when she would return. At some point she would get a message on her pager from a third party and start heading home [correction: it seems that only 12 experiments used the pager in this way, the rest were pre-planned return times]. Meanwhile at home there is a camera set up watching the spot the dog waits by the window when the owner is coming home. Later, someone looks at the tape and records how much time the dog spends by the window.
They break the data up into 3 different types, the return time starts the moment she gets the beep on the pager and lasts for 10 minutes (the trip home is always at least 13 minutes, so this 10 minutes will be time spent travelling home but not when the car is close enough for the dog to hear or smell). The pre-return time is the 10 minutes preceding the pager beep, and the main period is the time before that, broken up into 10 minute chunks.
The amount of time spent by the window was compared for the three different time periods, during the return period, the dog spent about 55% of it's time by the window, in the pre-return period he spent about 23%, and in the main period he spent about 4% of the time. (note: there is a lot more detail than this in the paper, and there are some nice charts, but this gives a pretty good flavor of it)
They also ran 10 experiments when the owner didn't return at all to rule out the possibility that the dog simply spent more time at the window the longer she was away. This is graphed in figure 5 in the paper. It seems to me that the dog spent a little more time at the window the later in the day, but not by enough to account for the result we saw in the other experiments.
Ruling out normal explanations
1. Routine, between doing random return times and different times of day, this can't be it
2. Hearing, Due to distance and alternate vehicles a few times, it can't be this
3. Reading people at home, they didn't know return time, can't be this
4. Selective memory blind recorders, can't be this
5. Spending time at window more later in the day control experiment, can't be this
I mostly agree with these points. There seems to be a little bit with the spending time at the window more and more as the day goes on though, it is important to keep it in mind, but by itself it doesn't explain the results.
In the section entitled "Thirty ordinary homecomings" he describes 5 ways that "interesting details are hidden by this averaging process". Again, I'm not a scientist and I have never designed a study like this, so maybe this type of thing is just the reality you have to deal with, but it seems to me that these things could have been taken care of with a little more care at the design stage. Perhaps I am wrong on this count, but it definitely seems like it might be important.
Sheldrake argues that the best explanation is that the dog and his owner have a telepathic link. As she is on her way home the dog gets excited and waits by the window. But then what about the pre-return period? The owner wouldn't know yet that she is about to head home, and yet the dog is already waiting by the window anticipating her coming home. Surely this is not supported by the telepathy hypothesis. There are 3 answers in the paper, the first two they agree seems far fetched, perhaps the dog is telepathically linked with the person sending the pager message. They solved this by having an alternate person send the message sometimes and that didn't stop the behavior. Another answer is that the dog is precognitive.
They mention those possibilities, but they don't seem to really think that is the answer, what they think is that the owner knows that the beep is coming during some predetermined range. Even though she hasn't gotten the beep, she is anticipating the beep and therefore the dog reads that through their telepathic link. This explains why the amount of time near the window increases a little bit in the pre-return period. She anticipates the beep, he gets a little excited and waits by the window, then when she gets the beep she is actually heading home and the dog gets really excited and waits by the window even more.
What do I think really happened?
The last thing I wrote is really telling. The owner was anticipating what was happening in the study and introduced noise into the data. Doesn't the fact that their study was set up in such a way that it could be anticipated show that there are flaws in the design? If the owner could anticipate things is it possible that the dog could as well? I think the answer to this is yes. Dogs are really intuitive and pick up on a number of cues we give them without even being aware of it ourselves, we basically bred them to have this trait. Perhaps the dog realizes that when the camera is set up she is coming back within 4 hours (the length of the tape). In the control experiment when she didn't come home perhaps she inadvertently indicated to the dog she wouldn't be back that day. Perhaps the dog realized that she never comes home within an hour and a half and didn't wait by the window for her during that period.
I'd like to add that if my ideas are correct about how the dog anticipated things, that is still really damn cool. If it is possible for the dog to pick up on a suite of minor cues from the owner and make a reasonable guess as to when she was going to get home, that is awesome! If there is something else going on, I would be interested in finding out exactly what it is. It certainly does seem like Sheldrake has found something here. I don't for a second think that it is telepathy, although I suppose if it is more experiments could show it.
Researcher Degrees of Freedom?
There is also the question of statistical manipulation. I am not accusing him of intentionally lying and trying to trick us into believing his story. I watched that youtube video of this guy and I think he really believes this stuff. But the fact remains that there are a lot of different ways to manipulate data (relevant xkcd comic). We don't know how much time he spent looking at different analyses before he decided this was what was best. Perhaps there are other ways to break up the data that would not show nearly as good of a result. We can't know if this happened, perhaps he decided exactly how the data would be analyzed ahead of time and it just worked out, I don't know, but it is something to be aware of. I will say that the way the 10 minute blocks are set up seemed strange to me and it really makes me wonder if this kind of thing was involved there. For instance, instead of averaging all of the main period, what if you only looked at the last 10 minute block in the main period? Maybe that is what they did already, maybe the original plan didn't have the 'pre-return' period but their data was coming out strangely and they put it in their to fix things. This is of course wild speculation but I think it is at least as probable as the dog and owner having a telepathic link :)