18/06/06

Sta$s$cs  and  Quality  Assurance

Discovery  of  inert  gas

Chemical  Nitrogen  (g) Nitrogen  from  Air  (g)
2.30143   2.31017
2.29890 2.30986
2.29816 2.31010
2.30182 2.31001
2.29869 2.31024
2.29940 2.31010
2.29849 2.31028
2.29889 Lord  Rayleigh
Average The  Novel  Prize  in  Physics  1904
2.29947 2.31011
Standard  devia$on
(Error  =  random  error)
0.00138 0.00014

Is  the  two  results  significantly  different?

2

1
 18/06/06

Mean  value  and  standard  devia$on

Mean  (average) Sample  standard  devia$on

∑x i ∑(x − x ) i
2

i
x= s= i
n n −1
n −1 :    the  degree  of  freedom
s  =  1
s2 :    the  variance
s  =  2 The  smaller  the  value  s,  the  more  closely  
the  data  are  clustered  about  the  mean.  
s  =  3
If n = ∞,

-­‐10   -­‐5   0   5   10  
x = µ, s =σ
3

Standard  devia$on  and  probability

Gaussian  curve

1 2 2 µ  =  0,  σ  =  1
y= e−( x−µ ) /2σ
0.4
σ 2π
Total  area  =  1.0

Normal  error  curve

x−µ x− x
z= ≈ area  =  0.0227
σ s
σ =1
0.054
1 −z2 /2
y= e -­‐4 0 2 4
2π
4

2
 18/06/06

Standard  devia$on  of  the  mean

Standard  devia$on  of  the  mean  sets  of  n  values
σ
σn =
n
Ex)  We  assume  that  infinity  data  are  included  in  the  bucket.  

mean = x1
σ
σ5 =
5
Five  data  
mean = x2
Five  data  
.  
.  
.  
n=∞ .  
.   mean [ x1, x2 , x3,⋅⋅⋅⋅] = µ
.  
mean = µ  

SD = σ
5

Confidence  limits  (CL)

Confidence  limits  define  a  numerical  interval  around  x  that  contains  µ  with  
a  certain  probability

CL  of  single  measurement

CL  =  50%  
-­‐0.67σ  ~  +0.67σ
CL = x ± zσ

CL  =  80%   CL  of  repeated  measurement  (N)

-­‐1.29σ  ~  +1.29σ

zσ
CL  =  95%  
CL for µ = x ±
-­‐1.96σ  ~  +1.96σ N

-­‐4σ -­‐2σ 0 2σ 4σ

3
 18/06/06

Confidence  intervals  (CI)

Finding  the  confidence  interval  when σ is  unknown
x−µ
t= “t”  is  the  desired  confidence  level
s
ts
CI = x ±
N
Values  of  Student’s  t
Confidence  level  (%)
Degrees  of  
freedom 80 90 95 99 99.9
1   3.08 6.31 12.7 63.7 637
2 1.89 2.92        4.30        9.92        31.6
3 1.64 2.35        3.18        5.84        12.9
4 1.53 2.13        2.78        4.60                8.60
5 1.48 2.02        2.57        4.03                6.86
6 1.44 1.94        2.45        3.71                5.96

How  to  calculate  the  CI

Nitrogen  from  Air  (g)
Confidence  level  =  80%
x = 2.31011 1.44 × 0.000143
CI = 2.31011±
s = 0.000143 7
N =7
= 2.31011± 0.000078
Degree of freedom = 6

2.30990 2.31000 2.31010 2.31020 2.31030

±  Standard  Devia$on

CI = 2.31011± 0.000078
80%  confidence  for  7  measurements

CI = 2.31011± 0.000132
95%  confidence  for  7  measurements
It  is  cri$cal  to  specify  what  kind  of    uncertainty  you  are  repor$ng.
8

4
 18/06/06

Comparing  replicate  measurements

Q:  Are  the  results  of  two  different  measurements  “within  experimental  error”?

Before  comparison
1.  Examining  an  outlier  (Grubbs  text)  
2.  Comparing  standard  devia$ons  (F  test)

2.30990 2.31000 2.31010 2.31020 2.31030

Is  this  outlier  (ques$onable  value  Q)?

Is  the  devia$on  significantly  different?

S1
S2

S3

Grubbs  test  for  outlier

2.30990 2.31000 2.31010 2.31020 2.31030

x = 2.31011
Is  this  outlier  (ques$onable  value  Q  =  2.30986)? s = 0.000143
N =7

Grubbs  test G  table

1.  Compute  the  average  and  standard  devia$on  (s)  of  the   Number  of   G    (95%  
complete  data  set  (N=7)   observa$on confidence)
2.  Compute  the  Grubbs  sta$s$c  G  with  the  following  eq. 4   1.463
Q−x 5 1.672
Gcalculated = 6 1.822
s
7 1.938
=  1.748 8 2.032
If  Gcalculated  is  greater  than  the  value  in  G-­‐table,  the  value   9 2.110
Q  should  be  discarded.

10

5
 18/06/06

Comparison  of  standard  devia$on  with  the  F  test

Is  the  devia$on  significantly  different?
S1
S2

S3

s12
Fcalculated = s1 > s2 (F ≥ 1)
s22
If  Fcalculated  >  Ftable,  the  difference  is  significant.
F  table
Degrees  of   Degrees  of  freedom  for  s1
freedom  s2 2 3 4 5 6 7
2 19.0 19.2 19.2 19.3 19.3 19.4
3 9.55 9.28 9.12 9.01 8.94 8.89
4 6.94 6.59 6.39 6.26 6.16 6.09
5 5.79 5.41 5.19 5.05 4.95 4.88
6 5.14 4.76 4.53 4.39 4.28 4.21

11

Comparing  replicate  measurements

Calcula$ng  Pooled  Standard  Devia$on
∑(x − x ) + ∑ (x
i 1
2
j − x 2 )2
s12 (n1 −1) + s22 (n2 −1)
set1 set 2
s pooled = =
n1 + n2 − 2 n1 + n2 − 2

n1:  the  number  of  results  in  set  1  

n2:  the  number  of  results  in  set  2  
n1+n2-2:  degrees  of  freedom

Then,  calcula$ng  the  value  of  tcalculated

x1 − x2 n1n2
tcalculated =
s pooled n1 + n2

If  tcalculated  >  ttable,  the  difference  is  significant.

12

6
 18/06/06

Comparing  replicate  measurements

When  the  devia$on  of  set  1  is  significantly  different  from  that  of  set  2,  how  to  
calculate  the  value  t?

x1 − x2
tcalculated =
s12 s22
+
n1 n2

2
! s12 s22 $
# + &
Degrees of freedom =
" n1 n2 %
2 2
! s12 $ ! s22 $
# & # &
" n1 % " n2 %
+
n1 −1 n2 −1

If  tcalculated  >  ttable,  the  difference  is  significant.

13

Case  study  of  Rayleigh’s  experiments

1.  Comparison  of  the  standard  devia$on
s12 0.001382
Fcalculated = =
s22 0.0001432
= 93.1 (> 4.21, n1=7, n2=6 )

2.  calcula$ng  the  value  of  tcalculated

2.31011− 2.29947
tcalculated = = 21.7
0.0001432 0.001382
+
7 8
Degrees of freedom = 7.17 (~7)

tcalculated = 21.7 > 5.048 (degrees of freedom = 7, confidence level = 99.9%)

The  difference  of  the  two  results  is  significant.

14

7
 18/06/06

Least  squares  methods

Vertical deviation (di) = yi – y = yi – (mxi + b)

Slope =
Δy
=m di2 = (yi − y)2 = (yi − mxi − b)2
6 Δx

5
Minimizing  the  squares  of  the  devia$ons
4 σy (xi, yi)
y ∑(x y ) ∑ x
i i i
3 m= ÷D
∑y n i
2 y = mx + b
1 ∑(x ) ∑(x y )
2
i i i
b= ÷D
∑x ∑y
i i
0 1 2 3 4 5 6
x
∑(x ) ∑ x
2
i i
D=
Finding  slope  (m)  and  intercept  (b)  with  a  spreadsheet  in  Excel ∑y n
i

15

Reliability  of  least-­‐squares  parameters

Popula$on  standard  devia$on  of  all  y  values
y = mx + b
6
σ y ≈ sy =
∑( d ) i
2

5 σy n−2
di
4 (xi, yi)
y Standard  devia$on  of  slope  m  and  intercept  b
3
2 sy2 n
2 s =
m
D
1
2
sy2 ∑ ( xi2 )
0 1 2 3 4 5 6 s =
b
x D
2

Uncertainty in x(= sx ) =
sy 1 1 (y − y)
+ + 2 i
m k n m ∑ ( xi − x ) 2

k:  the  number  of  replicate  measurements  of  an  unknown  sample

16

8
 18/06/06

Take-­‐home  message

Sta$s$cs  is  the  key  of  science.

17