Correlation of Sleep with Temperature and Seasons

[patrissimo]

I think you should probably see if there is any correlation after taking into account the total amount of sleep you get. You might perhaps be measuring a correlation between the time of the year and how much sleep you get. For me, a higher % in winter and a lower % in summer would mean I was sleeping in during the summer. You're not going to college right now, are you?

Right, maybe he gets more physical exercise when it's warmer, for example, I've heard that can increase deep sleep.

[zoocrewphoto]

Max: I also see in the more detailed recent data that there may be a leading of Deep vs. Temp, which may suggest you're sensitive to something changing (barometric pressure?).

I sleep better when it is cold in the bedroom. And this is better when the bedroom temperature is driven down by outside cold instead of A/C.

I sleep better during times of high barometric pressure.

Wonder why?

I don't know if my sleep is related to the barometric pressure, but I think my headaches are. I need to find a website where I can look up what the pressure was, but after the fact. If I know what it is and when it is changing, then that would affect my perception. But it would be nice to check them a few days later and see if my headaches correlate to changes in the pressure, and see if there are big changes that don't have headaches as well.

I do find that if I redo all my bedding so that I am really cozy warm, I sleep better. My heaviest comforter has a tendency to creep away from me and be totally off of me. I have tp work to pull it back across the bed. It is a feather down comforter and really heavy. I have also found that on a colder night, I sleep really well if I wear a long sleeve shirt. If I am cool, I tend to fuss a lot. But I have to be cold enough to wear the long sleeve shirt. I usually still have one bathroom trip, so I tend to change to a short sleeve shirt as a I am pretty warm by then. I am a severe night owl, so the heat is on long before I get up.

For me, it is finding the happy medium. Warm and cozy, but not hot. I love lots of heavy covers. I have a heated waterbed with 2 vellux blankets, the feather comforter, a regular comforter, and a knitted blanket on top. I also have another blanket within reach that I can use to cover my shoulders if I am still a little cold at the top. Also, I do not use sheets as they seem cold to me. I sleep on and under vellux blankets. When I travel in hotels, I take a heavy blanket with me and put it inside the sheets so that I am warm. If I travel with my sister, she likes the air conditioning, so I take a fleece jacket too. I always wear long leggings, all year round. And socks and slippers.

Years ago (over 10, maybe close to 20), I used to sleep in shorts and a tshirt, no socks. I can't imagine doing that now.

[Jay Aitchsee]

but my thought would be that you'd need a couple more variables to really bore down on this. If it were me, I'd look at sunlight exposure (both total and morning exposure) and outdoor physical activity as two more independent variables that can vary dramatically in winter time.

Presuming that there are groups/clumps within the data then it'd be interesting to determine the conditional correlation (the correlation within each group) to check if the 2 variables are still related once you factor out whatever might be causing the data to clump.

If I remember correctly, my deep sleep usually gets knocked out in the first few hours of sleeping, so my % of deep sleep would be lower on days I could sleep in because I only got (at most) a few hours of deep sleep no matter how long I slept. I would get a nice high total Z score on those days, though.
I think you should probably see if there is any correlation after taking into account the total amount of sleep you get. You might perhaps be measuring a correlation between the time of the year and how much sleep you get. For me, a higher % in winter and a lower % in summer would mean I was sleeping in during the summer. You're not going to college right now, are you?

Right, maybe he gets more physical exercise when it's warmer, for example, I've heard that can increase deep sleep.

Thanks all, for your inputs. Seth, I wish I were going to college right now. It's been quite a long time. I am aware that using Deep compared to Z probably skews the results some, but the relationship is similar using Deep alone. Noel and Old64, I think your right, there's more to this than a direct relationship to temperature. For example, Patrissimo's reference to exercise. In this case, because it's Florida, I get more exercise in the winter – which correlates.

Thinking about this, I imagine it quite possible that seasonal variations in Deep Sleep are the norm for most people. In fact, I think it is quite probable. Maybe some of you have your own data we could compare.

[Jay Aitchsee]

I don't know, here's the detail for Oct 1 through the 15th of January. The data is daily minutes of deep sleep vs daily high temperature. Both are smoothed by a 15 day moving average. An R squared of 0.80 seems pretty tough to argue with. Lending some credence to the relationship is the fact that since the first of the month or so my deep sleep has sharply declined, I haven't felt rested, and our temperatures have been unseasonably warm.

[SethW]

Statistics are a funny thing. Using a moving average as you are doing can invalidate assumptions used by some statistical tests. Plus the r-squared value does not necessarily mean the result is statistically significant.

Even if you did determine there was perfect correlation between the average temperature and the amount of deep sleep the Zeo says you get--then what? It's not as if you can adjust the outside temperature to change the amount of deep sleep you get. Besides, the correlation isn't perfect--but maybe it should be, and there is something you are doing that explains the difference! Identify that, and maybe you wouldn't be at the mercy of the daily high temperature or whatever.

What's the correlation between the outside temperature and the temperature where you sleep (which is presumably not outside)? WebMD says "experts agree the temperature of your sleeping area and how comfortable you feel in it affect how well and how long you snooze":

http://www.webmd.com/sleep-disorders/fe ... emperature

I don't have a lot of data in spite of having my Zeo for a year. I stopped using it once the sensors needed replacing. I just haven't got around to it.

[DreamStalker]

Have you compared your deep sleep data with hours of sunlight? Sunlight, specifically longer wavelengths caused by Mie scatter during low sun angles (ie. winter) and as the sun sets/rises is known to help convert serotonin hormone into melatonin hormone which in turn induces deep sleep. Plotting against incoming solar flux would be even better but I'm not sure of the data resolution for that.

Hours of sunlight would likely correlate to outside temperature. Hours of sunlight would be easier to explain with deep sleep than outdoor temperature.

I agree with Seth. What is your reason for using the moving average and why 15 days? -- Other than you just wanted to see what it does to the data.

One data transformation you could try is taking the ratio of deep sleep minutes to total sleep minutes. One could reason that if you only slept for 6 hours there would be less time available for deep sleep than say 9 hours. Of course if you always get the same amount of total sleep every night, such a transformation would be moot.

[Jay Aitchsee]

Right! I don't think I've found the secret to obtaining restorative sleep. However, there probably is a seasonal difference in the sleep of most people, likely influenced by some of the things you guys mentioned. I was surprised by the apparent correlation between the two variables, deep and temperature; although, I could probably find others that would seem to correlate also.

It's been a long time since I've studied statistics and I haven't had much call to to do this kind of data analysis over the years, but I find it interesting. There are probably lots of errors in my methods and assumptions, but it's still fun and it would appear some of you find it interesting and fun as well.

Anyway, I used the 15 day averages because there was just too much variability in the day to day numbers. Plotting those data points is not near as pretty Plus, the only temperature data I had for the first part of the year were the averages. Only in the last 4 months, or so, did I have each day's temperature. By using the the rolling average, I think I am showing the relationship in the trends of the variables over time. I am aware that the results could be completely artificial and even if there was a direct relationship between the outside temperature and deep sleep, so what? As Seth says, it's not like I can control the outside temperature - Well, I suppose I could move. Oh and DS, the first chart in this series does use percent of deep rather than minutes of deep and the results are similar. My total sleep time is fairly consistent.

Thanks all for your interest and comments.
Jay

[Jay Aitchsee]

Also surprising is that there appears to have been very few actual studies of the effects of seasons on sleep. On a quick trip through Google Land, I could find only one which seemed to address this question, and then the results were gathered by subjective self reporting by the participants. I did not find any where the results were gathered by objective means such as PSG.

The one study, Seasonal Variations in Sleep Problems at Latitude 63°–65° in Norway
The Nord-Trøndelag Health Study, 1995–1997, found little seasonal difference in the perceived quality of sleep.

http://aje.oxfordjournals.org/content/e ... wr052.full

http://sciencenordic.com/sleeping-well-summer-winter

[tbhit]

My two cents: I think that's a reasonably reliable correlation. I love sleeping when it's cooler out than when it's warm.

[Sleep2Die4]

I find it is surprising that someone thinks a correlation between lower temps and better sleep exists.

Even my wife, who doesn't like the cold, says she sleeps better when it is cold outside.

Is this not pretty much common knowledge?

[MaxDarkside]

Statistics are a funny thing. Using a moving average as you are doing can invalidate assumptions used by some statistical tests.

Yes, smoothing any two series will generally increase their correlation. Pearson's R also presumes linearity in the relationship between the two series (most often false in natural systems) and similar means and variances. So, in real life's chaotic non-linearity of significantly differently valued and ranging variables, while seeming useful, "correlation" actually has weak value other than a direction pointer for investigation.

Plus the r-squared value does not necessarily mean the result is statistically significant.

I'm not sure there's such a thing as "statistically significant correlation". Oh sure, "statisticians" think there is, but "statistically significant correlation" is to presume causation (the notion of actually being "significant") which is known to be false.

However, that said, what Jay is doing is valid because it gives him a "hey look!" and a path to investigate on a quest to at least understand his treatment and understand what other factors influence the benefits of same.

[DreamStalker]

Statistics are a funny thing. Using a moving average as you are doing can invalidate assumptions used by some statistical tests.

Yes, smoothing any two series will generally increase their correlation. Pearson's R also presumes linearity in the relationship between the two series (most often false in natural systems) and similar means and variances. So, in real life's chaotic non-linearity of significantly differently valued and ranging variables, while seeming useful, "correlation" actually has weak value other than a direction pointer for investigation.

Plus the r-squared value does not necessarily mean the result is statistically significant.

I'm not sure there's such a thing as "statistically significant correlation". Oh sure, "statisticians" think there is, but "statistically significant correlation" is to presume causation (the notion of actually being "significant") which is known to be false.

However, that said, what Jay is doing is valid because it gives him a "hey look!" and a path to investigate on a quest to at least understand his treatment and understand what other factors influence the benefits of same.

Yep. I think to determine the non-linearity Jay should plot the probability distributions of each series to check for non-linearity and if that is the case he could just try a log transformation of the non-linear series.

Sorry Jay --- not like you have a life to live or anything other than to please cpaptalk observers.

[Jay Aitchsee]

Yep. I think to determine the non-linearity Jay should plot the probability distributions of each series to check for non-linearity and if that is the case he could just try a log transformation of the non-linear series.

Sorry Jay --- not like you have a life to live or anything other than to please cpaptalk observers.

Sure, I'll get right on that

[MaxDarkside]

... Jay should plot the probability distributions of each series to check for non-linearity and if that is the case he could just try a log transformation of the non-linear series.

While log / exponential relationships are common in nature, they are but only one. Most "statisticians'" solution to handle non-linearity is to use "logistic regression", which may have no bearing whatsoever on the relationship. In fact, Series A does not even have to be correlated to Series B to be causative.

[DreamStalker]

... Jay should plot the probability distributions of each series to check for non-linearity and if that is the case he could just try a log transformation of the non-linear series.

While log / exponential relationships are common in nature, they are but only one. Most "statisticians'" solution to handle non-linearity is to use "logistic regression", which may have no bearing whatsoever on the relationship. In fact, Series A does not even have to be correlated to Series B to be causative.

Yep very true.

However as I noted earlier, the temperature data may be acting as a surrogate to sunlight. And sunlight (specifically long wavelengths of amber and red colors) is known to have a causal effect on melatonin production which has a causal effect on deep sleep. So we may have Series A (temperature) correlated non-causal surrogate to Series B (sunlight) correlated causal to Series C (melatonin) correlated causal to Series D (deep sleep) ... and of course each of those correlations is not likely to be linear. I still think that ideally, daily hours of sunlight should be used but that data would be hard to find. If there is a weather station or university nearby (sometime even the local TV weather stations), they may have daily or hourly solar radiation data.

Did you say you wanted to go back to school Jay? Here's your chance to go back with a thesis in hand.