bgy5901
TRANSCRIPT
DATA ANALYSIS
types of data/ variables
measures of central tendency
measures of dispersion, SE
assumptions of parametric tests
tests of normality
homogeneity of variance
hypothesis testing
T-test
ANOVA
Outline
Definition of Biostatistics
STATISTICS : Field of study relating to the collection, classification, summarization, analysis and interpretation of numerical information.
Definition of Statistics
BIOSTATISTICS : Application of statistics to the analysis of biological and medical data.
Experimental unit- object, person, anything upon which a treatment is applied.
A factor of an experiment is a controlled independent variable, experimenter determines levels of factor.
Level- different values of a factor. Level implies amount or magnitude.
Response variable is the dependent variable which is dependent on the factor.
Definitions
a complete set of units of interest
can be general or specific
usually determined by research question
parameter- any measure that tells us something about the entire population; uses lower case of Greek letters e.g. μ
Population
Sample a fraction of the population
cannot collect data from everyone in population- why?
statistic- any measure that tells us something about the sample; uses upper case of Latin letters e.g. X
use sample to estimate parameters of population
must be representative and random
sample should reflect the composition of the population of interest
every person (or unit) in the population from which the sample is drawn has an equal probability of being chosen
Sample as good estimators of population
representative
random
Descriptive vs. Inferential statistics
descriptive
inferential
the use of graphical or numerical methods to summarize and identify patterns in a data set
only provides information on data being analyzed
the use of sample data to make generalizations about a larger set of data
provides estimation about population of interest based on selected parts of the population
A variable is any measured characteristic or attribute that differs for different subjects.
What is a Variable?
For example, if the length of 45 leaves were measured, then length would be a variable.
Levels of Precision in Measurement
names assigned to categories but no relation between the categories can be inferred
Nominal
Ordinal
Interval
Ratio
values are ranked (put in order)
distance between any two adjacent values is the same but the zero point is arbitrary
similar to interval level but contains an absolute zero point
Types of data/ variables
Matric. No. Marks Position
123456 98 1
123457 95 2
123458 72 3
123459 71 4
123460 60 5
Example of ordinal scale
1 2 3 4 5 6 7 8 9 10
interval same length
Example of interval scale
Measures of central tendency
Central tendency is the point at which the distribution of
scores is centred.
Three measures of central tendency:
1. Mode
2. Median
3. Mean
Measures of central tendency
Mode
the most frequent value
for nominal data the mode is the only measure of central tendency
easy to calculate and understand
possible to have several modes in a data set may not always represent the data well and can change if a new
value is added
Measures of central tendency
Median
the middle value of a distribution when the values are arranged in numerical order; if even number of values, take the average of the two middle values
stable: relatively unaffected by extreme values & skewed distributions
can be used with ordinal, interval or ratio data
sampling fluctuations: likely to differ in samples from same population
can be misleading when comparing samples therefore less useful than the mean
Measures of central tendency
Mean (average)
sum of all values of a variable divided by the number of values
uses every value (no loss of information)
most accurate summary of the dataresistant to sampling variation (if several samples taken from the
same population, means likely the same)
can be influenced by extreme values (outliers) can only be used with continuous data
Measures of dispersion
Dispersion refers to the variability of values in a data set i.e. the extent to which a set of values differ
Measures of dispersion
Range
difference between the highest and the lowest value
easy to compute
outliers: easily influences by extreme values
based on only two of the observations and gives no idea of how the other observations are arranged between these two numbers
tends to increase as the size of the sample increases
Measures of dispersion
Interquartile Range
range of the middle 50% of values
less susceptible to outliers
uses only half of the data
Standard deviation
average difference between each value and the mean
measures the variability within the data set
how well the mean represents the data
uses every value
can be influenced by extreme values
Measures of dispersion
Sampling Distribution
9
8
7
4
3
2
1
56
μ = 10
Sample Mean
1 8
2 10
3 9
4 10
5 10
6 11
7 12
8 9
9 11
Distribution of the sample means
3
2
1
8 9 10 11 12
Sample mean
Freq
uenc
ySampling distribution of the sample means
How well does the sample represent the population?
If we want to know how well the mean represents the data we calculate the standard deviation of the mean.
Similarly, to estimate how accurate the sample represents the population we will calculate the standard deviation of the distribution of the sample means i.e. the Standard Error (SE).
SE = standard deviation of the population/ √ n
Since the SD of population is not known, SD of sample will be used instead
Standard Deviation vs. Standard Error of the Mean
(SEM)
When to use which?
1. Independent valuesvalue from one subject does not influence the value of another
2. Interval datadata should be measured at least at the interval level
3. Normally distributedbell-shaped; tests of normality should be conducted
4. Homogeneity of variancevariances should be the same throughout the data
Assumptions of Parametric Tests
Tests of normality
1. Skewness and kurtosis
2. Histogram and stem and leaf plot
3. Kolmogorov-Smirnov & Shapiro-Wilk
tests
4. Normal probability plot (Q-Q plot)
5. Box-plot
Skewness and Kurtosis
values of skewness and kurtosis should be zero in a normal distribution
values of skewness & kurtosis should be divided by their respective standard errors
look for values greater than 1.96; if > 1.96 then data is NOT normally distributed
Descriptives
161.52 .765
160.01
163.03
161.57
161.70
94.212
9.706
127
190
63
13
-.170 .191
.804 .380
Mean
Lower Bound
Upper Bound
95% ConfidenceInterval for Mean
5% Trimmed Mean
Median
Variance
Std. Deviation
Minimum
Maximum
Range
Interquartile Range
Skewness
Kurtosis
height (cm)Statistic Std. Error
Skewness and Kurtosis
3.503.002.502.001.501.000.50
length between internodes (cm)
60
50
40
30
20
10
0
Fre
qu
en
cy
Mean =2.175Std. Dev. =0.47512
N =745
Histogram
Kolmogorov Smirnov & Shapiro-Wilk
compare the scores in the sample to a normally distributed set of scores with the same mean and standard deviation.
if test is non-significant (p > 0.05) then distribution is not significantly different from a normal distribution therefore it is normally distributed
if p < 0.05 then distribution is significantly different from a normal distribution therefore it is NOT normally distributed.
Kolmogorov Smirnov & Shapiro-Wilk
Tests of Normality
.036 161 .200* .991 161 .393height (cm)Statistic df Sig. Statistic df Sig.
Kolmogorov-Smirnova
Shapiro-Wilk
This is a lower bound of the true significance.*.
Lilliefors Significance Correctiona.
4.03.53.02.52.01.51.00.5
Observed Value
2.5
0.0
-2.5
Ex
pe
cte
d N
orm
al
Normal Q-Q Plot of length between internodes (cm)
Box-Plot
Median should be in the middle of the box.
Outliers
values that are widely separated from the rest
Possible reasons for outliers:
measurement invalid (device not functioning, misrecorded value)
misclassified measurement- belongs to a population different from which the rest of sample was drawn
represents a rare or chance event
Homogeneity of Variance
Levene’s Test
Tests if variances in different groups are the same.
If significant (p< 0.05) variances are NOT equal.
If non-significant (p > 0.05) variances are equal.
Variance Ratio (VR)
Compare two or more groups.
Variance ratio = largest variance/ smallest variance
If VR < 2, homogeneity can be assumed.
Components of a hypothesis test
1. null hypothesis (Ho)
2. alternative hypothesis (Ha)
3. test statistic
4. reject or accept? p-value vs. significance level
Hypothesis Testing
A tentative explanation for an observation, phenomenon, or scientific problem that can be tested by further investigation.
Hypothesis TestingYou have some claim about the parameter and you want to see whether the data supports the claim or not
Hypothesis
Null hypothesis (Ho)
statement being tested in a statistical test usually the null hypothesis is a statement of no effect
or no difference
Alternative hypothesis (Ha)
experimental hypothesis- a hypothesis to be
considered as an alternative to the null hypothesis
Null and alternative hypothesis
Definition
Value used to decide whether or not the null hypothesis should be rejected in hypothesis testing
Sources of variation
In any experiment there are two basic sources of variation
1. systematic- variation due to experimental manipulation
2. unsystematic- due to random factors
Test Statistics
Need to calculate test statistic to find differences between samples
test statistic = systematic variation
unsystematic variation
Then need to calculate the probability of obtaining a value that large
Compare the amount of variance created by an experimental effect against amount of variance due to random factors- WHY?
Test Statistics
if experiment has had an effect we’d expect it to create more variance than random factors alone
the bigger the test statistic, the more unlikely it is to occur by chance; probability of them occurring by chance becomes smaller
when probability falls below a certain pre-determined value, accept that test statistic as large as it is because of experimental manipulation and not due to random factors
Test Statistics
significance level, α
probability that the test rejects the null hypothesis on the assumption that the null hypothesis is true
pre-determined value
p-value
probability that the test statistic be as large or larger than that actually observed by chance alone if the Ho is true
the smaller the p-value, the stronger is the evidence against Ho i.e. the observed result is unlikely to occur just by chance
p-value and significance level
Statistical Significance
– In statistics, a result is called significant if it is unlikely to have occurred by chance.
– "A statistically significant difference" simply means there is statistical evidence that there is a difference.
– However it does not mean the difference is necessarily large, important or meaningful.
– Means that observed effects are unlikely due to chance and results are reliable and likely to be repeatable
Two kinds of errors can be made in significance testing
1. a true null hypothesis can be incorrectly rejected (Type I)
conclusion drawn that the null hypothesis is false when in fact it is true
probability of Type I error (α) is the significance level
2. a false null hypothesis can be failed to be rejected (Type II error)
considered an error because fail to reject the null hypothesis correctly e.g. assuming no effect of treatment when there was
probability of Type II error is β
Type I and II errors
1 – β = power of a statistical test
Power of a statistical test is the ability of a study to find a significant difference if indeed one exists.
It is the probability that you will reject the null hypothesis when it is false
Power of a statistical test
T-test
To look for differences in mean between group of subjects from two different experimental conditions
experimental condition- the procedure that is varied in order to estimate a variable's effect
If the mean difference between groups is large, it could mean two things:
1. The two groups were taken from the same population but differ simply due to chance.
2. The two groups come from different populations. (If for example we have manipulated one of the groups then this is evidence that the experimental manipulation has caused the large difference between the groups).
Independent & dependent t-test
Independent t-test
two experimental conditions
different subjects were assigned to each condition i.e. each subject is tested under only one condition.
Dependent t-test (paired t-test)
two experimental conditions
same subjects took part in both conditions of the experiment
Group Statistics
292 2.0766 .46262 .02707
453 2.2384 .47277 .02221
type of fertilizerinorganic
organic
length betweeninternodes (cm)
N Mean Std. DeviationStd. Error
Mean
Independent Samples Test
.421 .517 -4.597 743 .000 -.16174 .03518Equal variancesassumed
length betweeninternodes (cm)
F Sig.
Levene's Testfor Equality of
Variances
t dfSig.
(2-tailed)Mean
DifferenceStd. ErrorDifference
t-test for Equality of Means
Independent t-test (SPSS output)
compares mean from three or more groups
need to first test the Ho that all group means are equal; Ha is that the group means differ
if the null hypothesis is rejected this means that the means of these group are not equal
need to conduct post-hoc test to determine which means significantly differ
ANOVA
One-way independent ANOVA only one independent variable- age group (independent) and height
(dependent) and different participants will be used in each condition
Two-way independent ANOVA two independent variables- age group and gender (independent) and
height (dependent) and different participants will be used in each condition
One-way repeated measures ANOVA only one independent variable- exercise type (independent) and blood
pressure level (dependent) and same participants will be used in all conditions
Types of ANOVA
Identify the experimental unit, factor(s), response variable, level(s) and most appropriate statistical test.
1. 240 chickens of four different breeds were randomly assigned to three different farms. After five weeks the weight of the chickens were measured.
2. A researcher wanted to investigate whether different types of fertilizer mixtures affect the growth of plants differently. 36 seeds were randomly assigned to two different types of fertilizer treatment. The height of each plant was measured after 3 weeks.