Instrument design
Essential concept behind the design
Bandit Thinkhamrop, Ph.D.(Statistics)
Department of Biostatistics and Demography
Faculty of Public Health
Khon Kaen University
Begin at the conclusion
3
4
Caution about biases
Selection bias
Information bias
Confounding bias
Research Design
-Prevent them
-Minimize them
Caution about biases
Selection bias (SB)
Information bias (IB)
Confounding bias (CB)
If data available:
SB & IB can be assessed
CB can be adjusted using
multivariable analysis
Sampling design
Please refer to IPDET Handbook Module 9
Types of Random Samples
– simple random samples
– stratified random samples
– multi-stage samples
– cluster samples
– combination random samples.
Summary of Random Sampling Process
1. Obtain a complete listing of the entire population
2. Assign each case a unique number.
3. Randomly select the sample using a random
numbers table.
4. When no numbered listing exists or is not
practical to create, use systematic random
sampling:
– make a random start
– select every nth case.
Questionnaire design
Design it with purpose, valid and reliable
Wording and layout are important
Question types
– Multiple choice (radio button)
– Multiple-item responses (checkbox)
– Open-ended (blank or text area)
Think aloud and improve the questionnaire
Prepare manual of operation
Pre-testing and improve them
Type of the study outcome: Key for
selecting appropriate statistical methods
Study outcome
– Dependent variable or response variable
– Focus on primary study outcome if there are
more
Type of the study outcome
– Continuous
– Categorical (dichotomous, polytomous, ordinal)
– Numerical (Poisson) count
– Even-free duration
Continuous outcome
Primary target of estimation:
– Mean (SD)
– Median (Min:Max)
– Correlation coefficient: r and ICC
Modeling:
– Linear regression
The model coefficient = Mean difference
– Quantile regression
The model coefficient = Median difference
Example:
– Outcome = Weight, BP, score of ?, level of ?, etc.
– RQ: Factors affecting birth weight
Categorical outcome
Primary target of estimation :
– Proportion or Risk
Modeling:
– Logistic regression
The model coefficient = Odds ratio (OR)
Example:
– Outcome = Disease (y/n), Dead(y/n),
cured(y/n), etc.
– RQ: Factors affecting low birth weight
Numerical (Poisson) count outcome
Primary target of estimation :
– Incidence rate (e.g., rate per person time)
Modeling:
– Poisson regression
The model coefficient = Incidence rate ratio (IRR)
Example:
– Outcome =
Total number of falls
Total time at risk of falling
– RQ: Factors affecting tooth elderly fall
Event-free duration outcome
Primary target of estimation :
– Median survival time
Modeling:
– Cox regression
The model coefficient = Hazard ratio (HR)
Example:
– Outcome = Overall survival, disease-free
survival, progression-free survival, etc.
– RQ: Factors affecting survival
The outcome determine statistics
Continuous
Mean
Median
Categorical
Proportion
(Prevalence
Or
Risk)
Linear Reg.
Count
Survival
Rate per “space”
Median survival
Risk of events at T(t)
Logistic Reg. Poisson Reg.
Cox Reg.
Statistics quantify errors for judgments
Parameter estimation
[95%CI]
Hypothesis testing
[P-value]
Sample
n = 25
X = 52
SD = 5
Population
Parameter estimation
[95%CI]
Hypothesis testing
[P-value]
Z = 2.58
Z = 1.96
Z = 1.64
SD
SE 
n
5
SE 
25
5
5
= 1
Z = 2.58
Z = 1.96
Z = 1.64
Sample
n = 25
X = 52
SD = 5
SE = 1
Population
Parameter estimation
[95%CI] :
52-1.96(1) to 52+1.96(1)
50.04 to 53.96
We are 95% confidence that the population mean would lie between 50.04 and 53.96
Sample
n = 25
X = 52
SD = 5
SE = 1
Population
Hypothesis testing
H0 :  = 55
HA :   55
Z = 55 – 52
1
3
52
-3SE
55
+3SE
Hypothesis testing
H0 :  = 55
HA :   55
Z = 55 – 52
3
P-value = 1-0.9973 = 0.0027
1
If the true mean in the population is 55, chance to obtain a sample mean of 52 or
more extreme is 0.0027.
P-value vs. 95%CI (1)
An example of a study with dichotomous outcome
A study compared cure rate between Drug A and Drug B
Setting:
Drug A = Alternative treatment
Drug B = Conventional treatment
Results:
Drug A: n1 = 50, Pa = 80%
Drug B: n2 = 50, Pb = 50%
Pa-Pb
= 30% (95%CI: 26% to 34%; P=0.001)
P-value vs. 95%CI (2)
Pa > Pb
Pb > Pa
Pa-Pb = 30% (95%CI: 26% to 34%; P< 0.05)
P-value vs. 95%CI (3)
Adapted from: Armitage, P. and Berry, G. Statistical methods in medical research. 3rd edition. Blackwell Scientific Publications, Oxford. 1994. page 99
Tips #6 (b)
P-value vs. 95%CI (4)
Adapted from: Armitage, P. and Berry, G. Statistical methods in medical research. 3rd edition. Blackwell Scientific Publications, Oxford. 1994. page 99
There were statistically
significant different
between the two groups.
Tips #6 (b)
P-value vs. 95%CI (5)
Adapted from: Armitage, P. and Berry, G. Statistical methods in medical research. 3rd edition. Blackwell Scientific Publications, Oxford. 1994. page 99
There were no
statistically significant
different between the
two groups.
P-value vs. 95%CI (4)
Save tips:
– Always report 95%CI with p-value, NOT report
solely p-value
– Always interpret based on the lower or upper
limit of the confidence interval, p-value can be
an optional
– Never interpret p-value > 0.05 as an indication
of no difference or no association, only the CI
can provide this message.
Q&A
Thank you