iRobot
Member of DD Central
Posts: 1,680
Likes: 2,477
|
Post by iRobot on Jul 25, 2020 16:03:58 GMT
What you appear to be talking about here is a time-series analysis... And yet you call it a linear regression (which it can't be, because we only have one variable here and regression analyses are for examining the relationship between multiple variables). After this, I'm afraid I really do fail to follow what you're apparently showing here. But maybe, indeed, my own doctoral training in stats isn't up to the level of yours . I only did my undergrad in Tabland, alas... slumming it in London for the postgrad. On the slopes everyday I bet! unless you mispelt Tabland of course. I suspect less Tyrol more Tyneside? (Or I'm completely off track!)
|
|
chris1200
Member of DD Central
Posts: 827
Likes: 508
|
Post by chris1200 on Jul 25, 2020 16:04:38 GMT
On the slopes everyday I bet! unless you mispelt Tabland of course. Ha! I was referring - somewhat self-disparagingly, I suppose - to the Oxford nickname for Cambridge (Cantabrigian -> Cantab -> Tabs) Alas the place is somewhat different from your experiences these days, sadly. And even since my own much more recently. Fun appears to have been banned at my old College
|
|
aju
Member of DD Central
Posts: 3,496
Likes: 923
|
Post by aju on Jul 25, 2020 16:06:12 GMT
On the slopes everyday I bet! unless you mispelt Tabland of course. I suspect less Tyrol more Tyneside? (Or I'm completely off track!) Why Aye Man!
|
|
iRobot
Member of DD Central
Posts: 1,680
Likes: 2,477
|
Post by iRobot on Jul 25, 2020 16:08:06 GMT
On the slopes everyday I bet! unless you mispelt Tabland of course. Ha! I was referring - somewhat self-disparagingly, I suppose - to the Oxford nickname for Cambridge (Cantabrigian -> Cantab -> Tabs) Alas the place is somewhat different from your experiences these days, sadly. And even since my own much more recently. Fun appears to have been banned at my old College Good quiz question: "Cambridge is one of the Cantabrigian universities, what is the the other".
|
|
chris1200
Member of DD Central
Posts: 827
Likes: 508
|
Post by chris1200 on Jul 25, 2020 16:08:13 GMT
On the slopes everyday I bet! unless you mispelt Tabland of course. I suspect less Tyrol more Tyneside? (Or I'm completely off track!) I'm afraid so! Although correct for one of my siblings, as it happens. Perhaps I was also being a little naughty and this was a test of credentials - but one that has now been ruined!
|
|
aju
Member of DD Central
Posts: 3,496
Likes: 923
|
Post by aju on Jul 25, 2020 16:10:44 GMT
On the slopes everyday I bet! unless you mispelt Tabland of course. Ha! I was referring - somewhat self-disparagingly, I suppose - to the Oxford nickname for Cambridge (Cantabrigian -> Cantab -> Tabs) Alas the place is somewhat different from your experiences these days, sadly. And even since my own much more recently. Fun appears to have been banned at my old College Oh a Rugby Joke, I worked with a few guys who played with some of those, not many were actually Uni people though, the club was just up the road from whence I worked.
|
|
chris1200
Member of DD Central
Posts: 827
Likes: 508
|
Post by chris1200 on Jul 25, 2020 16:12:00 GMT
Oh a Rugby Joke, I worked with a few guys who played with some of those, not many were actually Uni people though, the club was just up the road from whence I worked. Certainly not just a rugby thing! Just part of the general silly rivalry at any and all Varsity events. Anyway, we are veering far, far from the topic at hand. So I shall stop!
|
|
aju
Member of DD Central
Posts: 3,496
Likes: 923
|
Post by aju on Jul 25, 2020 16:25:54 GMT
Ha! I was referring - somewhat self-disparagingly, I suppose - to the Oxford nickname for Cambridge (Cantabrigian -> Cantab -> Tabs) Alas the place is somewhat different from your experiences these days, sadly. And even since my own much more recently. Fun appears to have been banned at my old College Good quiz question: "Cambridge is one of the Cantabrigian universities, what is the the other". So do you mean a certain place with a similar name in MassOfChewSetts, USA perhaps, I seem to recall someone I met at Kings Bar one night suggested this might be the case but it was a very long time ago.
|
|
|
Post by diversifier on Jul 25, 2020 17:59:58 GMT
t-test? Plot the graph of the 18 weekly figures. It’s reducing during the period before that under investigation, so it needs to be de-trended before one can use the raw data. Standard linear regression: linear trend is -£0.12m per week, Correlation value 0.8. De-trend the data by subtracting that overall gradient. Calculate the population standard deviation of the de-trended weekly figures: £0.45m Let’s exclude the data from weeks 1-7, as we are trying to detect a *second* step-down in RYI amount. The Null Hypothesis is that the true average “recent” weekly (last 3 weeks) is the same as the true average “previous” weekly (weeks 8-15) The linear mean of weeks 8-15 is £3.6m, and the std dev of that estimate is £0.45m/sqrt(8) = £0.16m Recent Weeks 16-18 mean £2.9m, and the std dev of that estimate is £0.26m The means differ by £0.7m, with a combined stddev of £0.3m. *Therefore, we reject the Null Hypothesis at the 2.3-sigma level, which corresponds to p<0.001* Does that help? Please don’t say “I can’t follow your calculations” if you’re going to claim that *I’m* statistically illiterate. It may help you to pause your response, if you know that I have a PhD in quantum physics from Cambridge, and a long senior engineering career in mathematical methods and algorithms. What you appear to be talking about here is a time-series analysis... And yet you call it a linear regression (which it can't be, because we only have one variable here and regression analyses are for examining the relationship between multiple variables). After this, I'm afraid I really do fail to follow what you're apparently showing here. But maybe, indeed, my own doctoral training in stats isn't up to the level of yours . I only did my undergrad in Tabland, alas... slumming it in London for the postgrad. X-axis is time, y-axis is RYI amount. X-axis can be equally-spaced (and in this case is), but does not have to be. Two variables.
|
|
|
Post by diversifier on Jul 25, 2020 18:07:12 GMT
Ha! I was referring - somewhat self-disparagingly, I suppose - to the Oxford nickname for Cambridge (Cantabrigian -> Cantab -> Tabs) Alas the place is somewhat different from your experiences these days, sadly. And even since my own much more recently. Fun appears to have been banned at my old College Good quiz question: "Cambridge is one of the Cantabrigian universities, what is the the other". ahem....the other three! You’ll be meaning AHEC....aka Anglia Poly....aka Anglia Poly Uni....aka Anglia Ruskin But there’s also University of the Third Age, plus a branch of the Open University....
|
|
|
Post by ruralres66 on Jul 25, 2020 18:22:08 GMT
Seems to have strayed a tad from 'Current processing times for investment releases (RS)'......! Must be the stresses of Lockdown?
Nothing wrong with Anglia Ruskin University........
I got a first class Hons BA there as a mature student in 1998. (Education)
|
|
|
Post by stevepn on Jul 25, 2020 20:09:38 GMT
Seems to have strayed a tad from 'Current processing times for investment releases (RS)'......! Must be the stresses of Lockdown?
Nothing wrong with Anglia Ruskin University........
I got a first class Hons BA there as a mature student in 1998. (Education)
It may have been caused by our resident Financial Advisor.
|
|
|
Post by RateSetter on Jul 27, 2020 18:33:06 GMT
Good evening all. Today we have delivered £0.6m and the full update is below:
|
|
|
Post by bernythedolt on Jul 28, 2020 3:34:23 GMT
diversifier , sorry to do this, but as long as you keep making these assertions with such definitive confidence that you're correct, I'm going to say something. Contrast with my posts above illustrating my uncertainty and that I'm piecing bits of evidence together without being sure. Your post is completely statistically illiterate. - Nothing you specify is 'statistically significant'. Statistical significance allows us to reject a null hypothesis at a certain level of confidence based on a statistical test (e.g. t-test, z-test etc.). You seem to just be using it as a meaningless buzzword. - "We can now be sure that the average is below £3m per week" - what average? The average of the last two weeks: yes of course, that's pretty obvious. The average of all weeks so far: absolutely not. Again, what are you even saying here? - How can you know the level of reinvestment so confidently without knowing the levels of loan repayment and levels of new lending for this month? (Just for starters) [Edit: You've now added some calculations. I'm afraid that I fail to follow these also.] t-test? Plot the graph of the 18 weekly figures. It’s reducing during the period before that under investigation, so it needs to be de-trended before one can use the raw data. Standard linear regression: linear trend is -£0.12m per week, Correlation value 0.8. De-trend the data by subtracting that overall gradient. Calculate the population standard deviation of the de-trended weekly figures: £0.45m Let’s exclude the data from weeks 1-7, as we are trying to detect a *second* step-down in RYI amount. The Null Hypothesis is that the true average “recent” weekly (last 3 weeks) is the same as the true average “previous” weekly (weeks 8-15) The linear mean of weeks 8-15 is £3.6m, and the std dev of that estimate is £0.45m/sqrt(8) = £0.16m Recent Weeks 16-18 mean £2.9m, and the std dev of that estimate is £0.26m The means differ by £0.7m, with a combined stddev of £0.3m. *Therefore, we reject the Null Hypothesis at the 2.3-sigma level, which corresponds to p<0.001* Does that help? Please don’t say “I can’t follow your calculations” if you’re going to claim that *I’m* statistically illiterate. It may help you to pause your response, if you know that I have a PhD in quantum physics from Cambridge, and a long senior engineering career in mathematical methods and algorithms. Like chris1200 , I’d have (rightly or wrongly) approached this using Student's t-test on the null hypothesis, because of the low sample sizes (with the caveat that the sample variances are rather too dissimilar to provide an accurate test). The unbiased sample std dev of the weeks 8-15 data set (3.6, 4.0, 3.4, 3.5, 3.7, 3.5, 2.8, 4.3) is s 1 = 0.44 and for the weeks 16-18 data (3.6, 2.4, 2.7), s 2 = 0.62. Using the weighted average, with n 1 = 8 and n 2 = 3, I calculate the pooled std dev, s p = sqrt ( ( (n 1-1) s 12 + (n 2-1) s 22 ) / (n 1 +n 2 - 2) ) to be 0.49 Means differ by 0.7, so the t statistic is given by t = 0.7 / ( s p * sqrt (1/n 1 + 1/n 2) = 2.12 At 9 degrees of freedom, this is not significant at the 95% level (p=0.063), so I’d have concluded this isn’t quite sufficient to reject the null hypothesis! I’m looking at the data, you the standard error of their means, but why have we got such different outcomes? My degree was well over 30 years ago and I’m well rusty now, so please be gentle if I’ve made a gaffe but I’m intrigued why the two assessments lead to different outcomes, mine a borderline acceptance of H 0 but yours a definite rejection. How do we tell which is right? Could it be that looking at the data tells us more about their variance, which perhaps SEM loses sight of? Or perhaps the sample variances are too dissimilar, rendering the t-test too inaccurate to be of use. At one time I probably could have resolved this for myself, but these days I’m well out of my depth…
|
|
aju
Member of DD Central
Posts: 3,496
Likes: 923
|
Post by aju on Jul 28, 2020 7:43:59 GMT
t-test? Plot the graph of the 18 weekly figures. It’s reducing during the period before that under investigation, so it needs to be de-trended before one can use the raw data. Standard linear regression: linear trend is -£0.12m per week, Correlation value 0.8. De-trend the data by subtracting that overall gradient. Calculate the population standard deviation of the de-trended weekly figures: £0.45m Let’s exclude the data from weeks 1-7, as we are trying to detect a *second* step-down in RYI amount. The Null Hypothesis is that the true average “recent” weekly (last 3 weeks) is the same as the true average “previous” weekly (weeks 8-15) The linear mean of weeks 8-15 is £3.6m, and the std dev of that estimate is £0.45m/sqrt(8) = £0.16m Recent Weeks 16-18 mean £2.9m, and the std dev of that estimate is £0.26m The means differ by £0.7m, with a combined stddev of £0.3m. *Therefore, we reject the Null Hypothesis at the 2.3-sigma level, which corresponds to p<0.001* Does that help? Please don’t say “I can’t follow your calculations” if you’re going to claim that *I’m* statistically illiterate. It may help you to pause your response, if you know that I have a PhD in quantum physics from Cambridge, and a long senior engineering career in mathematical methods and algorithms. Like chris1200 , I’d have (rightly or wrongly) approached this using Student's t-test on the null hypothesis, because of the low sample sizes (with the caveat that the sample variances are rather too dissimilar to provide an accurate test). The unbiased sample std dev of the weeks 8-15 data set (3.6, 4.0, 3.4, 3.5, 3.7, 3.5, 2.8, 4.3) is s 1 = 0.44 and for the weeks 16-18 data (3.6, 2.4, 2.7), s 2 = 0.62. Using the weighted average, with n 1 = 8 and n 2 = 3, I calculate the pooled std dev, s p = sqrt ( ( (n 1-1) s 12 + (n 2-1) s 22 ) / (n 1 +n 2 - 2) ) to be 0.49 Means differ by 0.7, so the t statistic is given by t = 0.7 / ( s p * sqrt (1/n 1 + 1/n 2) = 2.12 At 9 degrees of freedom, this is not significant at the 95% level (p=0.063), so I’d have concluded this isn’t quite sufficient to reject the null hypothesis! I’m looking at the data, you the standard error of their means, but why have we got such different outcomes? My degree was well over 30 years ago and I’m well rusty now, so please be gentle if I’ve made a gaffe but I’m intrigued why the two assessments lead to different outcomes, mine a borderline acceptance of H 0 but yours a definite rejection. How do we tell which is right? Could it be that looking at the data tells us more about their variance, which perhaps SEM loses sight of? Or perhaps the sample variances are too dissimilar, rendering the t-test too inaccurate to be of use. At one time I probably could have resolved this for myself, but these days I’m well out of my depth… Considering I was never in the pool then it follows that I haven't a clue about these levels of musings/maths, mispent youth and all that, but for me the obvious thing for us is we just sit back and eventually it will get to me. Even if someone manages to find the errors/flaws/etc in these theories at the end of the day I am still at the back of the queue unless RS were to change tack and share the sales around ala Zopa which I feel is highly unlikely. The one thing I always keep in mind is "Lies, Damned Lies, and Statistics!" and the worse scenario has to be making things fit what we want them to fit. The other interesting thing might be what the next monthly report suggests, assuming there will be one and its not seriously curtailed/changed or other changes.
|
|