Teshale Sori

 We have our senses, to experience

the world & make observations

 We have the ability to reason,

enables us to make logical inferences

Statistics: science
of drawing conclusions from data
10/22/23
3

10/22/23
4

10/22/23
5

10/22/23
 Chance
6

10/22/23
 Chance
7
8

Example: Pigs from one abattoir; one

large farm 10/22/23
9

10/22/23
10

10/22/23
11

10/22/23
12

10/22/23
13

10/22/23
14

10/22/23
 Statistical testing
15

 The principle: to compare two or more means

 Take into account the variances & sample size
 Associates independent & dependent variables
 Independent: measurable with accuracy
 Eg: serum antibody vs disease
 Disease vs production
 Dependent variables: the variables of interest in the
observational study or the experiment
 production,
 mortality,
 parasite burden, . . .
 Example: milk production
16
. ttest milk, by(Ration)

Two-sample t test with equal variances

Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval]

basic 30 799.3667 26.9389 147.5505 744.2704 854.4629

suppl 30 994.3 38.34775 210.0393 915.87 1072.73

combined 60 896.8333 26.47206 205.0517 843.8629 949.8038

diff -194.9333 46.86421 -288.7422 -101.1244

diff = mean(basic) - mean(suppl) t = -4.1595

Ho: diff = 0 degrees of freedom = 58

Ha: diff < 0 Ha: diff != 0 Ha: diff > 0

Pr(T < t) = 0.0001 Pr(|T| > |t|) = 0.0001 Pr(T > t) = 0.9999
10/22/23
 . regress milk i.Ration

Source SS df MS Number of obs = 60

17
F( 1, 58) = 17.30
Model 569985.067 1 569985.067 Prob > F = 0.0001
Residual 1910741.27 58 32943.8149 R-squared = 0.2298
Adj R-squared = 0.2165
Total 2480726.33 59 42046.209 Root MSE = 181.5

milk Coef. Std. Err. t P>|t| [95% Conf. Interval]

Ration
suppl 194.9333 46.86421 4.16 0.000 101.1244 288.7422
_cons 799.3667 33.138 24.12 0.000 733.0337 865.6996
10/22/23
 Hypothesis testing
18

10/22/23
 Hypothesis testing
19

10/22/23
 Hypothesis testing
20
 Analysis of continuous data:
Linear Regression
21

 Learning outcomes
 Identify if the data is analyzed by
Linear regression
 Construct a linear regression model
 Interpret the linear regression
coefficients from both technical and
causal perspectives
 Assess the fitness of linear
regression model

10/22/23
 Analysis of continous variables:
Linear regression
22

Used to estimate one variable from the other

(assymetric)
10/22/23
23

 Example: predict head circumference of children

 Interested in probability distribution of HC of
population of low birth weight children (<1500gm)
o Mean = 27cm
o SD = 2.5cm
 Since the measurement is rougly normal, 95% of
the infants have HC between (22.1, 31.9)
 Suppose HC increases with gestational age
 Look at HC of infants by taking specific
10/22/23
gestational age
24

10/22/23
25

10/22/23
26

10/22/23
27 10/22/23
 y = 3x + 2 baseline prediction = 4
28

SST=(mean –
observed)^2

SST = 24
SSE=SUM(predicted
– observed )^2
SSM = 18

R^2 = 1 –SSM/SST

RMS=SSM/N^1/2
10/22/23
 Assumptions
29

For analysis:
 Linearity
 Normal distribution of the outcome
variable
For inference
 Normality of residuals
 Homeoscedasticity
10/22/23
 Simple Linear Regression: carcass price
30

10/22/23

Does the volume of sales (independent variable) inuence

the price per unit (dependent variable)?
 Regression
31

The t-test formula is a typical example of what we call a

statistical test

The ratio between the difference of two averages and its

standard error

The probability to obtain a value equal to or larger than this

ratio under the so-called null hypothesis is determined

i.e true difference between the two averages is zero

10/22/23
 Regression In Stata:

32
Source SS df MS Number of obs = 10
F( 1, 8) = 7.99
Model 34.1342349 1 34.1342349 Prob > F = 0.0223
Residual 34.1867667 8 4.27334584 R-squared = 0.4996
Adj R-squared = 0.4371
Total 68.3210017 9 7.59122241 Root MSE = 2.0672

pricecarcass Coef. Std. Err. t P>|t| [95% Conf. Interval]

numberofpigssold1000000 -.366877 .1298104 -2.83 0.022 -.6662202 -.0675337

_cons 45.55915 9.783604 4.66 0.002 22.99812 68.12018

P = 45.55915 -.366877*number
10/22/23
 Regression
33

10/22/23
34

10/22/23
 Multivariable regression : relations among t-test,
ANOVA and linear regression
35

10/22/23
36

 Questions to be answered:
 Does supplementation have significant
effect on the weight gain?
 Is growth significantly different among
breeds?
 Is the effect of the supplement different
among breeds?
10/22/23
 Source SS df MS Number of obs = 24
F( 7, 16) = 40.84
Model 11114.625 7 1587.80357 Prob > F = 0.0000
37 Residual 622 16 38.875 R-squared = 0.9470
Adj R-squared = 0.9238
Total 11736.625 23 510.288043 Root MSE = 6.235

adwg Coef. Std. Err. t P>|t| [95% Conf. Interval]

breed
nh 45 5.090841 8.84 0.000 34.2079 55.7921
lw 54 5.090841 10.61 0.000 43.2079 64.7921
p 37 5.090841 7.27 0.000 26.2079 47.7921

ration
s 21 5.090841 4.13 0.001 10.2079 31.7921

breed#ration
nh#s -4 7.199537 -0.56 0.586 -19.26234 11.26234
lw#s -5 7.199537 -0.69 0.497 -20.26234 10.26234
p#s -4 7.199537 -0.56 0.586 -19.26234 11.26234

_cons 200 3.599769 55.56 0.000 192.3688 207.6312

10/22/23
 Source SS df MS Number of obs = 24
F( 4, 19) = 81.80
Model 11092.5 4 2773.125 Prob > F = 0.0000
38 Residual 644.125 19 33.9013158 R-squared = 0.9451
Adj R-squared = 0.9336
Total 11736.625 23 510.288043 Root MSE = 5.8225

adwg Coef. Std. Err. t P>|t| [95% Conf. Interval]

breed
nh 43 3.361612 12.79 0.000 35.96406 50.03594
lw 51.5 3.361612 15.32 0.000 44.46406 58.53594
p 35 3.361612 10.41 0.000 27.96406 42.03594

ration
s 17.75 2.377019 7.47 0.000 12.77484 22.72516
_cons 201.625 2.657588 75.87 0.000 196.0626 207.1874
10/22/23
39

10/22/23
40

10/22/23
41

 1. Test of normality of residuals

 Before any other test, the normal

distribution of the residual values
must be confirmed

 Stata offers two commands to this

end:

 Shapiro-Wilk test and Shapiro-

Francia test
10/22/23
42

10/22/23
43

 2. Residual values in function

of predicted values

 A graph of the residual values in

function of the predicted values is
produced with the command
rvfplot (residual-versus-fitted
plot)

10/22/23
44

10/22/23
45

 The command hettest (heteroscedasticity test) tests the

equality of the variances of the residual values

10/22/23
46
Source SS df MS Number of obs = 308
F( 7, 300) = 78.06
Model 6.35654316 7 .908077595 Prob > F = 0.0000
Residual 3.48991243 300 .011633041 R-squared = 0.6456
Adj R-squared = 0.6373
Total 9.8464556 307 .032073145 Root MSE = .10786

temp Coef. Std. Err. t P>|t| [95% Conf. Interval]

co2 .0062906 .0025726 2.45 0.015 .001228 .0113532

ch4 -.0004803 .0005816 -0.83 0.410 -.0016248 .0006642
n2o -.0248186 .0091634 -2.71 0.007 -.0428512 -.006786
cfc11 -.0080571 .0017144 -4.70 0.000 -.0114309 -.0046833
cfc12 .005394 .0010219 5.28 0.000 .0033829 .0074051
tsi .0726611 .0162568 4.47 0.000 .0406693 .1046529
aerosols -.8517611 .2332851 -3.65 0.000 -1.310843 -.3926787
_cons -93.3341 22.22649 -4.20 0.000 -137.0737 -49.59453

10/22/23
47

mei co2 ch4 n2o cfc11 cfc12 tsi

mei 1.0000
co2 -0.1529 1.0000
ch4 -0.1056 0.8723 1.0000
n2o -0.1624 0.9811 0.8944 1.0000
cfc11 0.0882 0.4013 0.7135 0.4122 1.0000
cfc12 -0.0398 0.8232 0.9582 0.8393 0.8314 1.0000
tsi -0.0768 0.0179 0.1463 0.0399 0.2846 0.1893 1.0000
aerosols 0.3524 -0.3693 -0.2904 -0.3535 -0.0323 -0.2438 0.0832
temp 0.1353 0.7485 0.6997 0.7432 0.3801 0.6889 0.1822

10/22/23
48

 Multicollinearity arises when independent

variables are highly correlated among
themselves
 The effect of multicollinearity to mask real
effects of one or more independent variables
 Either preventing the null hypothesis to be
rejected or even by reversing the sign of the
regression coefficient

10/22/23
49

10/22/23
50

10/22/23
51

 The results of this analysis are highly

contradictory:
 On one hand a very high R2 value is
obtained & a highly significant F-value,
 On the other hand, none of the
regression coefficients are different
from zero
 Such a result is almost always a very
good indication of the presence of
multicollinearity,


10/22/23
52

10/22/23
53

10/22/23
54

10/22/23
55

10/22/23
56

10/22/23