January 2023 Mid-Long Range Disco

[psuhoffman]

16 minutes ago, WesternFringe said:

This is the link for my DC annual snowfall statistical analysis since 1888.  

Snowfall is barely decreasing when measured from 1880s (-.07” per year) and 1960s (-.03” per year).

Snowfall is up (+0.17” per year) when measured from the 1980s.

Most of the variability is random and statistical noise (96+%).

Doom and gloom, in my opinion, is just human error (recency effect) when interpreting recent winters emotionally.

These were my conclusions from the link above if you don’t want to view the link:

Conclusions:

Annual snowfall In DC has declined on average 0.07" per year since 1888.

Annual snowfall In DC has declined on average 0.03" per year since 1969.

Annual snowfall In DC has increased on average 0.17" per year since 1984.

The vast majority of the variability (94% up to 99.8%, depending on the time period observed) from year to year is statistical noise, or random, and not due to the passage of time.

A mean of such a low number can be skewed by a few years. I’d be more interested in a statistical analysis of the median and changes in the standard deviation. 

[87storms]

15 minutes ago, psuhoffman said:

A mean of such a low number can be skewed by a few years. I’d be more interested in a statistical analysis of the median and changes in the standard deviation. 

I ran DCA stats over the weekend for fun with a few starting points of 1887, 1950 and 1990.  Just eyeballing from those datasets, it looks like the median is slowly decreasing and stdv is slowing increasing.  If correct, that would make some sense and align with the idea of less stat padding storms and more hit/miss, larger precip events.  WesternFringe can probably add more to this.

[WesternFringe]

20 minutes ago, psuhoffman said:

A mean of such a low number can be skewed by a few years. I’d be more interested in a statistical analysis of the median and changes in the standard deviation. 

n = 136 isn’t such a low number, but I can run a statistical analysis of the median of decades.  
 

I think mean is more useful over 136 years than medians of decades, which are a man-made concept, but I will take a look.

[DarkSharkWX]

 

[dallen7908]

The problem with the mean is that the distribution is becoming increasingly skewed because the mean is low and bounded by zero.  N=136 is a healthy sample size. 

 

[WesternFringe]

25 minutes ago, WesternFringe said:

n = 136 isn’t such a low number, but I can run a statistical analysis of the median of decades.  
 

I think mean is more useful over 136 years than medians of decades, which are a man-made concept, but I will take a look.

Especially when I start to look at trends since the 1960s.  My n = 62 will go to n = 6

[WesternFringe]

8 minutes ago, dallen7908 said:

The problem with the mean is that the distribution is becoming increasingly skewed because the mean is low and bounded by zero.  N=136 is a healthy sample size. 

 

Usually studies with n=18 and higher start to become highly generalizable to the general population and those below it do not (thus my issue with looking at decade median data, especially from recent decades). Even looking at decadal data from 1880s only gives us n=14.  That isn’t a generalizable sample size.

At least that is what I was taught at University of Virginia when I was getting my doctorate

eta:  I think looking at the slope of the standard deviation from 1880s might be useful.  Definitely more useful than median decadal data

[mattie g]

Nerds

[pazzo83]

23 minutes ago, WesternFringe said:

Usually studies with n=18 and higher start to become highly generalizable to the general population and those below it do not (thus my issue with looking at decade median data, especially from recent decades). Even looking at decadal data from 1880s only gives us n=14.  That isn’t a generalizable sample size.

At least that is what I was taught at University of Virginia when I was getting my doctorate

eta:  I think looking at the slope of the standard deviation from 1880s might be useful.  Definitely more useful than median decadal data

yep this is what I was always taught - you need a sample size of around 30.  I think it's interesting the median is decreasing but the std deviation is increasing.  That would suggest that while overall snowfall is trending down just a bit, we are dealing with more all or nothing type scenarios (which would make sense if we are dealing with an underlying state of bigger storms but a warming atmosphere (meaning generally less snow here)).

[WesternFringe]

1 hour ago, psuhoffman said:

A mean of such a low number can be skewed by a few years. I’d be more interested in a statistical analysis of the median and changes in the standard deviation. 

Also, if you had a decade of

3,4,4,4,4,5,8,20,25,30,40,45 (like we seem to do with Ninas), is 5 inches or does 17.5” encapsulate the decade snowfall better?  Not sure, just thinking aloud.

[WesternFringe]

11 minutes ago, pazzo83 said:

yep this is what I was always taught - you need a sample size of around 30.  I think it's interesting the median is decreasing but the std deviation is increasing.  That would suggest that while overall snowfall is trending down just a bit, we are dealing with more all or nothing type scenarios (which would make sense if we are dealing with an underlying state of bigger storms but a warming atmosphere (meaning generally less snow here)).

But again, given that you need a sample size of 18+ (30?), how can you say what the median decreasing and the standard deviation even means, when the n = 14?

[pazzo83]

Just now, WesternFringe said:

But again, given that you need a sample size of 18+ (30?), how can you say what the median decreasing and the standard deviation even means, when the n = 14?

yeah it's tough.  Even going on a year interval, DC snowfall has a ton of year-to-year variance given its location.  Another interesting thing to look at would be to see the frequency of 3"+, 6"+ and 12"+ events per year.

[WesternFringe]

My whole point is not to be a downer, but the opposite.  Stats say these bad years recently are aberrations and that we should have some good years coming soon!  Like, good, big years!!

[pazzo83]

47 minutes ago, dallen7908 said:

The problem with the mean is that the distribution is becoming increasingly skewed because the mean is low and bounded by zero.  N=136 is a healthy sample size. 

 

that's a great point - I guess you'd model it on some sort of distribution related to the exponential (beta maybe?)? 

[pazzo83]

2 minutes ago, WesternFringe said:

My whole point is not to be a downer, but the opposite.  Stats say these bad years recently are aberrations and that we should have some good years coming soon!  Like, good, big years!!

i mean there is some statistical validity to saying "we're due" - absolutely.

[WesternFringe]

6 minutes ago, pazzo83 said:

that's a great point - I guess you'd model it on some sort of distribution related to the exponential (beta maybe?)? 

Just treat it normally.  We don’t treat height that way, even though it is bound by zero and the mean is low.

eta: in fact, all observable occurrence data is bound by zero.  That doesn’t mean we model it’s distribution differently 

[pazzo83]

7 minutes ago, WesternFringe said:

Just treat it normally.  We don’t treat height that way, even though it is bound by zero and the mean is low.

eta: in fact, all observable occurrence data is bound by zero.  That doesn’t mean we model it’s distribution differently 

that's true - once your sample size is big enough.

[WesternFringe]

18 minutes ago, pazzo83 said:

that's true - once your sample size is big enough.

Right.  And 136 is big enough 

ETA:  I will exit stage left at this point.  I am just saying, I don’t believe the doomsayers in here with regard to future snowfall.

DCA is getting more snow than they were in the 1980s.

The data show the rate of change since 1880s is low and negligible and likely due to statistical noise, but the human brain likes to find patterns, even when there aren’t any

[eurojosh]

Not to belabor the statistics, but I'd be curious to see an analysis of how much snow in each season was a result of 'big' storms (either as a function of amount - say 6"+ or as a function of percent - say, 33% of seasonal snowfall) as opposed to spread out over multiple smaller events; my sense is that climate change reduces our overall odds of frozen precip and bunches what does fall into increasingly atypical large events - but I've no idea if the data would bear that out or not.

[WesternFringe]

4 minutes ago, eurojosh said:

Not to belabor the statistics, but I'd be curious to see an analysis of how much snow in each season was a result of 'big' storms (either as a function of amount - say 6"+ or as a function of percent - say, 33% of seasonal snowfall) as opposed to spread out over multiple smaller events; my sense is that climate change reduces our overall odds of frozen precip and bunches what does fall into increasingly atypical large events - but I've no idea if the data would bear that out or not.

Yeah, that is what we have been talking about.   Would love to see someone show me that scenario with numbers, and not a “we should see larger events” explanation.  Show me the data.

[Ji]

Please talk about weather

[WesternFringe]

Actually tired of hearing we are failing bc of climate change and not that we are failing (we haven’t failed yet for 22-23) bc of la nina which we all know to be a real thing.

eta:  obviously, climate change is a real thing, bc the climate has always been changing 

[WesternFringe]

8 minutes ago, Ji said:

Please talk about weather

I was responding to others, but yes Ji, we can get get back to the ‘fact’ that this winter is over bc you have no blue on your models.  Forgive me.

[WesternFringe]

We can suck with annual snowfall.  And we can suck with almost no snowfall some years.  But no one has shown me statistically that we are sucking more than we used to suck.  We have always sucked.  Prove me wrong, with stats.  Tired of hearing the negative Nancies.

I grew up in upstate NY where we averaged 65+”

ETA:  love Bob Chill, bc he is always grounded

[Its a Breeze]

17 minutes ago, WesternFringe said:

Actually tired of hearing we are failing bc of climate change and not that we are failing (we haven’t failed yet for 22-23) bc of la nina which we all know to be a real thing.

Not sure climate change cracks to the top 10 reasons why we fail, yet...

We're neither north nor south, stuck between mountains and an ocean. I think that's reason enough for boom or bust....

[psuhoffman]

1 hour ago, 87storms said:

I ran DCA stats over the weekend for fun with a few starting points of 1887, 1950 and 1990.  Just eyeballing from those datasets, it looks like the median is slowly decreasing and stdv is slowing increasing.  If correct, that would make some sense and align with the idea of less stat padding storms and more hit/miss, larger precip events.  WesternFringe can probably add more to this.

This is what I suspect. 

1 hour ago, WesternFringe said:

n = 136 isn’t such a low number, but I can run a statistical analysis of the median of decades.  
 

I think mean is more useful over 136 years than medians of decades, which are a man-made concept, but I will take a look.

I meant the mean is low and bounded by 0 like Dallen said not that the sample size is low.  I think median gives a much better idea of what a "typical" winter was like during any given period of time.  For example the last 16 years (this is the period of time where I've really noticed the phenomenon I am hinting at here) there have been 9 single digit snowfall seasons, 4 years between 10-20", 1 season between 20-30" and 2 BIG years above 30" that are skewing the mean.  During that time the mean is over 13" but the median is around 8".  8" being much closer to what a typical winter has been in DC over that period since 9 years were single digit snowfall and only 7 were more than that. 

1 hour ago, dallen7908 said:

The problem with the mean is that the distribution is becoming increasingly skewed because the mean is low and bounded by zero.  N=136 is a healthy sample size. 

 

Thank you, you said what I meant but much better. 

1 hour ago, WesternFringe said:

Usually studies with n=18 and higher start to become highly generalizable to the general population and those below it do not (thus my issue with looking at decade median data, especially from recent decades). Even looking at decadal data from 1880s only gives us n=14.  That isn’t a generalizable sample size.

At least that is what I was taught at University of Virginia when I was getting my doctorate

eta:  I think looking at the slope of the standard deviation from 1880s might be useful.  Definitely more useful than median decadal data

You're totally right, we need a larger sample size than 10 so I wouldn't use a 10 year median.  But you could do a running 20 or 30 year median.  But something is going on that is not refutable which is the prevalence of single digit snowfall seasons in DC is skyrocketing lately.   Looking at the data there is no 16 year period that even comes close to having 9 single digit seasons.  16 years is starting to become a bit much to continue the "its just recency bias" or "its just a short term blip" mantras.  And it's not like its a little off...its way off.   I know 16 years is kinda arbitrary but going back over all previous 16 year periods DC had 5, 4, 3, 5, 5, 3, and 3 single digit snowfall seasons....yet during the most recent 16 years there were 9.  Even if you try to simply cherry pick the worst 16 year periods you can't get close.  The worst you can find cherry picking is another 16 year period with 6 unless you overlap the current period and then you can find 7 single digit seasons, but that is only by including several of the seasons in this current 16 year period.  Basically the current most recent 16 years have been the absolute worst in recorded history in DC for getting abysmal low snowfall seasons...and the next worst is a 16 year period overlapping this current 16 year period!  And nothing else comes close.  That seems significant to me, more significant than to dismiss with "just random chance".  The mean doesn't show it as much because a few seasons have been skewing the number enough to hide what I am talking about...which is getting more frequent dreg single digit snowfall seasons .

Now what we don't know is this....is this awful stretch just part of a random cyclical thing or a more permanent shift.  It's very possible that if records went back further we could find similar awful periods...and perhaps we pull out of this and in the next 16 years only have 3 or 4 single digit seasons (but do you really wanna bet on that???).  So I am NOT saying I know this is all due to warming (although I suspect some of it is) but what I am saying is we are in a pretty long term stretch where its more likely than in any other period in recorded history for DC to have single digit snowfall year to year.  It's been a long enough period now that until we break out of it...I don't consider it just a blip.  Its not been just a few years...this trend has been going on for quite a while now with just a few interruptions by some big seasons here and there.  But other than those big seasons....suddenly DC's typical winter is a single digit snowfall...when that used to relatively rare in the past. 

[psuhoffman]

33 minutes ago, Ji said:

Please talk about weather

Is it partly cloudy or partly sunny? 

[ravensrule]

@stormtracker home videos are way more exciting than the past 2 pages. Can you please post some Randy, so i don't have to poke my eyeballs out from what i have just read. 

[WesternFringe]

2 minutes ago, psuhoffman said:

This is what I suspect. 

I meant the mean is low and bounded by 0 like Dallen said not that the sample size is low.  I think median gives a much better idea of what a "typical" winter was like during any given period of time.  For example the last 16 years (this is the period of time where I've really noticed the phenomenon I am hinting at here) there have been 9 single digit snowfall seasons, 4 years between 10-20", 1 season between 20-30" and 2 BIG years above 30" that are skewing the mean.  During that time the mean is over 13" but the median is around 8".  8" being much closer to what a typical winter has been in DC over that period since 9 years were single digit snowfall and only 7 were more than that. 

Thank you, you said what I meant but much better. 

You're totally right, we need a larger sample size than 10 so I wouldn't use a 10 year median.  But you could do a running 20 or 30 year median.  But something is going on that is not refutable which is the prevalence of single digit snowfall seasons in DC is skyrocketing lately.   Looking at the data there is no 16 year period that even comes close to having 9 single digit seasons.  16 years is starting to become a bit much to continue the "its just recency bias" or "its just a short term blip" mantras.  And it's not like its a little off...its way off.   I know 16 years is kinda arbitrary but going back over all previous 16 year periods DC had 5, 4, 3, 5, 5, 3, and 3 single digit snowfall seasons....yet during the most recent 16 years there were 9.  Even if you try to simply cherry pick the worst 16 year periods you can't get close.  The worst you can find cherry picking is another 16 year period with 6 unless you overlap the current period and then you can find 7 single digit seasons, but that is only by including several of the seasons in this current 16 year period.  Basically the current most recent 16 years have been the absolute worst in recorded history in DC for getting abysmal low snowfall seasons...and the next worst is a 16 year period overlapping this current 16 year period!  And nothing else comes close.  That seems significant to me, more significant than to dismiss with "just random chance".  The mean doesn't show it as much because a few seasons have been skewing the number enough to hide what I am talking about...which is getting more frequent dreg single digit snowfall seasons .

Now what we don't know is this....is this awful stretch just part of a random cyclical thing or a more permanent shift.  It's very possible that if records went back further we could find similar awful periods...and perhaps we pull out of this and in the next 16 years only have 3 or 4 single digit seasons (but do you really wanna bet on that???).  So I am NOT saying I know this is all due to warming (although I suspect some of it is) but what I am saying is we are in a pretty long term stretch where its more likely than in any other period in recorded history for DC to have single digit snowfall year to year.  It's been a long enough period now that until we break out of it...I don't consider it just a blip.  Its not been just a few years...this trend has been going on for quite a while now with just a few interruptions by some big seasons here and there.  But other than those big seasons....suddenly DC's typical winter is a single digit snowfall...when that used to relatively rare in the past. 

You are kind of proving my point by looking so intensively at the last 30 years as if they represent the data set on whole.  A running 10 to 20 year data set?  It is almost like you want to massage your numbers to show a point.  What is wrong with 136 yrs of  data points unless it doesn’t show what you want it to show? Lol

[WesternFringe]

Just now, WesternFringe said:

You are kind of proving my point by looking so intensively at the last 30 years as if they represent the data set on whole.  A running 10 to 20 year data set?  It is almost like you want to massage your numbers to show a point.  What is wrong with 136 yrs of  data points unless it doesn’t show what you want it to show? Lol

But tell me annual snowfall is decreasing at DCA since the 1980s and I can call you a liar and back it up with data