Anda di halaman 1dari 1

24 Analysing geochenvcal data

Talk 2.! Values for the correlation coefficient (r) above which it is Considered iuriuiciily significant
(it the 1 % or 5 % level) for I given number of degrees of freedom (DF)*

D Two-sided D Two-sided
F test One-sided test F test One-sided test
5% 1% S% 1% 5 1% 5 1%
% %
0.9 099 0. 0.3 0.4S 0.3
1 97 9 93 0.959 25 81 7 23 0.445
2 0.9 0.99 0. 0.930 30 0.3 0 0.2 0.409
3 50
o.m 0
0.95 90
0. 0.934 35 49
0,3 449
0.41 96
0.2 0.381
4 031 9
0.91 83
0. 0.882 40 25
0.3 S
0.39 75
0.2 0.358
5 1
0.7 7
0.87 72
0. 0.833 50 04
0.2 3
0.35 57
0.2 0.322
6 S4
0.7 5083 66
0. 0.789 60 73
0.2 40.32 31
0.2 0295
7 07
0.6 40.79 62
0. 0.7 70 50
0.2 50.30 11
0.1 0.274
8 66
0.6 3
0,76 58
0. SO
0.715 0 32
C.2 2
0.28 95
0.1 0.257
)2
0.6 5
0.73 54
0. 0.635 S 17
0.2 3
0.26 83
0.1 0.242
-1 02
0.5 5
0.70 S2
0. 0.658 O
10 05
0.1 7
0.25 73
0.1 0.230
01 76
0.S 80.66 49
0. 0612 150 9S
0.1 40.20 64
0.1 0.1S9
21 32
0.4 10.62 45
0. 0.S74 260 59
0.1 30.18 34
Q.1 0.164
14 77
0.4 3
0.59 42
0. 0543 300 38
0.1 01 16
0.0 0.134
16 68
0.4 0
0.56 40
0. 0.S1 400 13
0.0 .48
0.12 95
0.0 0.116
23 44
0.4 1
0.S3 37
0. 6
0.492 500 98
0.0 8
O.lt 82
0.0 6.10*
0 23 7 36 0 33 S 74
Dau fiom Sachs (19S4). 3

The two-sided test is used when p 4 0. For ctample, the data-3 in Tsblc 2.2 contains 31 samples and the calculated
correlation coefficient between CaO and AJJOJ IS 0.568 (Table 2.3a). Inspection of tabulated values for r (one-sided
test) shows that at the 5% significance level and 29 degrees of ficedom (n-2) the tabulated value for r is 0.301. Since
the calculated value (0.S68) is greater than the tabulated value (0.301), the correlation coefficient in tle sample is
statistically significant at the S% level. That is, there is 95% chance that the relationship observed in the sample also
applies to the population. Haicc, the null hypothesis that p = 0 is rejected.

2.3.3 Assumptions In the calculation of the product-moment coefficient of correlation

The Pearson product-momcnc coefficient of linear correlation b based upon the following
assumptions:

(1) The units of measurement are equidistant for both variables.


(2) There is a linear relationship between the variables.
(3) Doth variables should be normally or nearly normally distributed.

Assumption (3) is frequently regarded as an important prerequisite for linear correlation (e.g. Till,
1574). However, this is not always practised (Sachs, i984i for strictly it b the variation of y from the
estimated value of y for each value of x that must be normally distibutcd and rarely is the sample
population isrge enough for this criterion to be satisfactorily tested. Nevertheless, when tciting for the
significance of r, the data should be normally distributed.

Anda mungkin juga menyukai