Can you spot the trend in this data?

[snowsure]

Most people would admit under oath that they probably look at the preferred outcome of a data set before drawing their conclusions. So, thought I, why not ask people's opinion of trends in a graph without declaring what the data set is drawn from?

I will declare that it looks at changes in temperature (degree Celsius) from one data entry to the next. I will not declare at this stage what the time interval is between data entries.

This may be uncomfortable for some of you as it could, perhaps, expose you as a sympathiser for "the other side."

I will post, on Thursday 23rd, the source of the data.

Please tell me if you think that the megatrend is positive, negative or if there is no obvious trend.

[Bobby]

I can't really pick out any real trend to be honest.

[Candice]

As an analyst I would say that the megatrend is just about positive.

[Bobby]

There are methods to mathematically determine correlation, so what people perceive it to be is really irrelevant from a scientific standpoint.

[gmoran]

There are methods to mathematically determine correlation, so what people perceive it to be is really irrelevant from a scientific standpoint.

looks like a slight warming

[Gray-Wolf]

I think you're a trickster and it is a 'sunshine plot' for Formby last wednesday........

[snowsure]

I think you're a trickster and it is a 'sunshine plot' for Formby last wednesday........

You scallywag, you.

The timescale is appropriate for the discussion of climate change (IMO) and the region covered as well as the source of the data will stand be classed, by most, as acceptable.

Does Formby have such a "wild" variation of sunshine over a day/week/month/season/year/decade/century...

[Candice]

There are methods to mathematically determine correlation, so what people perceive it to be is really irrelevant from a scientific standpoint.

Yes - but isn't that what he's trying to show - it's about perception not mathematics

[Candice]

CET - 30 yrs ?????????????????????????????????????

[BARRY]

looks pretty neutral to me

[pottyprof]

Overall, I'll say neutral.

There are methods to mathematically determine correlation, so what people perceive it to be is really irrelevant from a scientific standpoint.

Not true. Perception like this could show some pretty good information from a scientific standpoint. Not sure what the data is but it could be from a poll and shows how people perceive global warming. Then again it could show melting rate of chocolate... :unsure:

[androcles]

OK. It's late. And as a 'sometimes' trader I would short this - depending on the timescale.

This is going South - long term.

Andy

[davehsug]

No obvious trend to me. I suppose the line seems to spend more time in the positive, so maybe I'd go for that if really pushed.

Dave

[Thundery wintry showers]

My reading:

A very slight positive trend, but nowhere near significant enough to be able to deduce "it's getting warmer". It has a strongly 'statistically insignificant' look about it.

Edited August 21, 2007 by Thundery wintry showers

[HighPressure]

Yep there is more above the line then below so slightly positive, say 0.5-0.75 using my eyes?

[noggin]

I have looked at the graph, but studiously avoided looking at what other posters have posted (to avoid any influence, IYKWIM).

The graph is temperature, if I understand correctly?

On the basis that it is temperature, I am plumping for it being the CET for the last 100 years and that the only trend is of ups and downs.....neither up nor down being predominant.

Wonder how far off the mark I am? :o !

[Anti-Mild]

I would say it was the rolling CET for the last 100 years with the 3 negative 'blips' nearest the right being the 80's, 60's and 40's.

[snowsure]

Thanks for taking part in my little experiment. There are some obvious notable exceptions on this link; people who are not comfortable working with "unknown" data perhaps?

The source of the data is the CET (well done Candice!) from 1974 to 2006. http://www.metoffice.gov.uk/research/hadle.../HadCET_act.txt

The graph depicts annual variances i.e. the first data point is the difference between 1975 and 1974. The next is the difference between 1976 and 1975, etc. This may not be the standard way of looking for a correlation. However if I want to know if I am going to run out of money I look at how much money I had this month, last month, etc and see where the trend is taking me.

Results are :

+ve 5 people

-ve 1 person

No correlation 3 people

Statistically insignificant as TWS says in my opinion.

Once again, thanks those of you brave enough to have a go.

SS

[Candice]

And my prize is ?

[Anti-Mild]

Forgive me if i'm being a little thick, I'm not a statistician, but the graph doesn't really show anything other than that temperatures vary year after year in a non-linear manner.

I suggest that if you took any 30 year period from anywhere in the CET series and recorded the data in same way then the line would follow a similar track.

Again, sorry if I'm misinterpreting but that's all I get from it.

[snowsure]

Forgive me if i'm being a little thick, I'm not a statistician, but the graph doesn't really show anything other than that temperatures vary year after year in a non-linear manner.

I suggest that if you took any 30 year period from anywhere in the CET series and recorded the data in same way then the line would follow a similar track.

Again, sorry if I'm misinterpreting but that's all I get from it.

I am at the same place as you AM (i.e I am not a statistician.) Looking for a pattern is not meant to be difficult.

If the climate was getting warmer I would expect there to be a preponderance of temperature increases year on year. As such, I do not feel that warming is yet happening in this linear manner.

I am sure that I could select a 30 year period with either a +ve or a -ve correlation if I wanted to but that proves nothing (except that warming/cooling has occured in the past and will do so again.)

The aim of the graph was to show people that looking for a pattern without knowing what the data shows is outside most peoples comfort zone. For example, you said what you thought the data was, not what the likely trend was.

However, I am sure a statistician will damn me for employing such spoil tactics. This was the equivalent to blind-tasting a wine. I understand peoples reluctance to get involved.

[Stratos Ferric]

I am at the same place as you AM (i.e I am not a statistician.) Looking for a pattern is not meant to be difficult.

If the climate was getting warmer I would expect there to be a preponderance of temperature increases year on year. As such, I do not feel that warming is yet happening in this linear manner.

I am sure that I could select a 30 year period with either a +ve or a -ve correlation if I wanted to but that proves nothing (except that warming/cooling has occured in the past and will do so again.)

The aim of the graph was to show people that looking for a pattern without knowing what the data shows is outside most peoples comfort zone. For example, you said what you thought the data was, not what the likely trend was.

However, I am sure a statistician will damn me for employing such spoil tactics. This was the equivalent to blind-tasting a wine. I understand peoples reluctance to get involved.

Sorry SS, only just found this.

You're certainly correct to say it's not a normal way of looking for a correlation - and that's because as such, it would be hugely invalid if what you're seeking to show is overall trend in temperature. I could draw you a whole host of number sets all of which would have positive trend but which could include one year of warming and 29 of cooling. Correlations test an independent variable and a dependent variable; what you've drawn is effectively a time series, where any year's value is dependent on that of the year before (since we know climate exists within fairly narrow bounds when we have had a big dip the likelihood is that we will then see an upswing, and v.v.).

The acid test, in the series you've given, of whether or not there's warming or cooling is to sum the value of each point: what you will then have is the sum change in temperature across the entire series, and THAT (and that alone) will be the indicator of trend. For any data set starting at zero, where the sum of the point as you've derived them is positive, the overall trend in the data will always be of the same sign as the sum itself. The more interesting question, derived differently, is whether or not the relationship is significant, and, in any case, irrespective of any degree of correlation, whether or not there is genuine cause and effect at work.

[snowsure]

... Correlations test an independent variable and a dependent variable; what you've drawn is effectively a time series, where any year's value is dependent on that of the year before (since we know climate exists within fairly narrow bounds when we have had a big dip the likelihood is that we will then see an upswing, and v.v.)

Hi SF

As you know, I like learned responses on this forum even when I have difficulty understanding them! This is not a dig at you but an ackowledgement of my inability to understand to your depth of knowledge. I am but a simple bloke!

So, to come to your first point. If we have 12 months of decreasing temperatures and they were shown on the same time series could it be possible to use a large enough rolling average to "hide" the data? Say a 10 year rolling average over 50 years? What about the last 2 months CET? Does their lack of statistical significance make the likelihoods of a new trend invalid?

The acid test, in the series you've given, of whether or not there's warming or cooling is to sum the value of each point: what you will then have is the sum change in temperature across the entire series, and THAT (and that alone) will be the indicator of trend. For any data set starting at zero, where the sum of the point as you've derived them is positive, the overall trend in the data will always be of the same sign as the sum itself. The more interesting question, derived differently, is whether or not the relationship is significant, and, in any case, irrespective of any degree of correlation, whether or not there is genuine cause and effect at work.

Is it a statistical suicide to look at the, say, 1900 to 1960 CET record and then try to predict the post 1960 results? A model that can accurately predict the previous 47 years will possibly be able to predict the next 5, 10 or 20 years. Has this been done?

Once again SF, thanks for your input. You do not go down in my estimation even though your CET punt for August was, shall we say, wildly wrong!

[Stratos Ferric]

Hi SF

As you know, I like learned responses on this forum even when I have difficulty understanding them! This is not a dig at you but an ackowledgement of my inability to understand to your depth of knowledge. I am but a simple bloke!

So, to come to your first point. If we have 12 months of decreasing temperatures and they were shown on the same time series could it be possible to use a large enough rolling average to "hide" the data? Say a 10 year rolling average over 50 years? What about the last 2 months CET? Does their lack of statistical significance make the likelihoods of a new trend invalid?

Is it a statistical suicide to look at the, say, 1900 to 1960 CET record and then try to predict the post 1960 results? A model that can accurately predict the previous 47 years will possibly be able to predict the next 5, 10 or 20 years. Has this been done?

Once again SF, thanks for your input. You do not go down in my estimation even though your CET punt for August was, shall we say, wildly wrong!

Last things first - wasn't it, though I think there were one or two even further out than me. To be honest, I've been expecting what we've now got: a long settled spell. We shall see how September pans out.

Re projection on the basis of the past, this is one of the tools that I use. However, and as Tamara is apt to point out, there is no certainty that says you can extrapolate from the past into the future. If you were to do that then a carve up of the CET from, say, 1900 to around 1940 would suggest continued warming - what then followed was thirty or so years of cooling. Similarly, take that thirty years and project onwards past the late 70s, and we'd be headed for a mini ice-age: what have we had in practice? Record warming. Therefore, a couple more factors need to be taken into account: one is the known bounds within which a variable oscillates; the other is corroborating data. Taking these in turn. Our climate tends to bounce around about 1C (roughly) either side of the running thirty year average. Therefore - assuming that no other fundamental forcing is at work - when a short trend tends to move towards the top of the typical variability you could reasonably expect a move back towards the middle. Hence why continued warming this year would have been unlikely, and a significant increase even had there been more warming could have been discounted - the caveat above notwithstanding. Gong back to your plot, therefore, it would be unlikely to see consecutive years of big movement in the same direction - what I WOULD expect is cycles up and down, and I suspect that if I recheck your numbers that is what I'll see.

Moving on to your point about time series length. There are statistical tests of robustness of hypothesis e.g. if I take a string of numbers that appear to have a rising trend, how confident can I bee that this trend isn't just random? As a rule, the larger the data set, the more confidence I can have in any calculation of probability, though once I get beyond 30, and certainly 60 or so, any increases in "robustness" become very marginal indeed. Two data points, therefore, are meaningless on their own. That said, as part of a longer series they become more relevant. AS I have already said elsewhere, the significance of the two months just past will only really become apparent in 12 months' time, if not longer.

If I get chance tomorrow evening I'll post some more figures to exemplify some of these points.

[snowsure]

If I get chance tomorrow evening I'll post some more figures to exemplify some of these points.

Thanks for imparting your knowledge there SF. I am certain that I am not the only one who will benefit from it. I must admit that I take an all too simplistic view on climate. This web site certainly helps folk like me to learn more.

I look forward to your examples whenever they become forthcoming.