finding things out

problems:

x y

seeing through randomness knowing when you have

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

15

solutions
x
seeing through randomness: large numbers averaging knowing when you have measuring variability statistical significance

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

16

large numbers
really random 25 drops too uniform

200 drops

more drops less difference

Avoiding Damned Lies Understanding Statistical Ideas, Alan Dix www.meandeviation.com 17

summing
add up several races some random effects cancel overall difference: large: 83-76 = 7 proportionate difference: small: 7/83 < 10% average scores close: 8.37.6 = 0.7 heads
5 10 8 10 10 9 10 10 6 5

tails
10 7 10 4 7 10 3 5 10 10

83

76

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

18

fairness and independence

fairness: each outcome has same probability
i.e. probability of head = 1/2

independence: each toss has same probability

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

19

fairness affects average

number of heads n prob(head) but not exact . . . . . . the world is very random

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

20

unbiased coin
10 series of 20 tosses, prob(head)=0.5:
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: TTTHHHTHHHTHHHTTHTHH HTTTTHTHHHTHHTTHHTHH HHTTTTTHHTTHHTHHTHTT THTTTHHHHHTHTHHTHHHH HTTTTHTTTHHTHHHHTTTT TTTTHTTTHHHHTTHTTTHH TTTTHHHTTHTHTHTHHTTT THHHTHTTHHTHTHHTHTTH HTTTTTTTTHHTTTHTHHHH HHHTTHHHHTTTHHTHTTTH 12 11 9 13 8 8 8 11 8 11

average

9.9

= 1.9

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

21

biased coin
prob(head) = 0.8:
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: HHHHHHHHHHTHHHHHHHTH THTHHTHHHHHHHHHHHHHH THHHTHTHHHHHHHHHHHHT HTHHHHHHHHHTHHHHTHHH HHHTHHHHHHHHHHHHTHHH HHHTHHHHTHHTHTHHHHHT HTTTHTHTHHHHHTHTHHHH HHTTHHTHHHHTHHTTTHTH HHHHHHHHHHHHHHHHTHHH HTHHHHHHHHHHHHHTHHHH 18 17 16 17 18 15 13 12 19 18

average

16.3

= 2.3

Avoiding Damned Lies Understanding Statistical Ideas,

Alan Dix

www.meandeviation.com

22

independence affects variability

independence: context doesn't matter
e.g. prob( head after head ) = prob( head after tail )

positive correlation: things vary together

e.g. prob( head after head ) > prob( head after tail )

negative correlation: things vary in opposite way

e.g. prob( head after head ) < prob( head after tail )
Avoiding Damned Lies Understanding Statistical Ideas, Alan Dix www.meandeviation.com 23

positive correlation
p(H) = p(T), but p(H after H) > p(H after T)
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: HTTTTTTTTTTTTTTTTTTT TTHHHHHTHHHHHHTHHTTT HHHHHTTHHHHHHTTHTTHH HTHHTTHHHHHHHHHHHHTT HHHHHHTHHHTTTTTTTTTT TTTHTTTTTHTHHHTTHHHH TTTTTTTTTHTTHTTTTTTT HHHHHTTTTTTTTTTTTTHH TTTTTHTHHHHHHHHTHTHH HHHHHHHHTTTHHHHTTHTH 1 13 14 15 9 9 2 7 12 14

average

9.6

= 5.0

long runs of heads and tails high variability of head count

Avoiding Damned Lies Understanding Statistical Ideas, Alan Dix www.meandeviation.com 24

negative correlation
p(H) = p(T), but p(H after H) < p(H after T)
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: TTHTHHTHHTHTHTHTTHTH TTHHTHTHTHTTHTHHTHHT HTHTTHHTTHTHTHTHTHTH THTHHTTHTHTTHTHTHTTH HTHTHTHTHTHHTTHHTTHT TTHTHHTTHHTHTHTHTHTH TTHHTTHTHTHTHTHTHTHH HTHTHTTHTHHTTHTTHHHT HHHTHHHTHTHTTHHHTHHT THHHHTHTTHTTHHTTHTHT 10 10 10 9 10 10 10 10 13 10

average

10.2

= 1.0

alternating heads and tails low variability of head count

Avoiding Damned Lies Understanding Statistical Ideas, Alan Dix www.meandeviation.com 25