A 40-minute plenary lecture which was addressed to Filipino educators. Lecture focused on five major issues, namely: Knowledge Sharing, sample size, statistical modeling, old school way of teaching statistics in the graduate school, and teaching statistics using statistical software.
Teaching Correct Statistical Methods in the Era of Knowledge Sharing
1. Teaching Correct Statistical Methods
in the Era of Knowledge Sharing
John T. Amora
Associate Professor of Statistics and Analytics
Head, Office of the Performance Assessment, De La Salle-College of Saint Benilde, Manila
Chairman/President, PARSSU
Bataan Research Educators Organization, Inc. (BREO)
INTERNATIONAL RESEARCH CONFERENCE
April 7-9, 2018 || Colegio de San Juan de Letran – Bataan, PHILIPPINES
2. Agenda
I. Knowledge Sharing
II. Sample Size Issues
III. Statistical Modeling Issues
IV. Updates on Statistical Models
V. Teaching Statistics in the Graduate School:
Manual Computation
VI. Teaching Statistics using Software
3. I. Knowledge Sharing/Dissemination of
Research Results
Ways of Disseminating Research Outputs?
Some Colleges/Universities (School A):
Research outputs are just kept in the
cabinet or placed in the book shelves.
Only few can read.
Researches not available and accessible
online.
Not aware of predatory journals,
predatory conference organizers.
4. I. Knowledge Sharing/Dissemination of
Research Results
Ways of Disseminating Research Outputs?
Other Colleges/Universities (School B):
One of their goals is to generate new knowledge.
Their outputs are published in the genuine and quality
journals (e.g., ISI, Scopus).
5. I. Knowledge Sharing/Dissemination of
Research Results
How do you Disseminate your Research Outputs?
Other Colleges/Universities (School B):
Research outputs are made available online (e.g.,
researchgate, google scholars, academia).
Aware of predatory journals and predatory conference
organizers.
Don’t plagiarize.
6. II. Sample Size Issues
a. Misuse of Slovin’s Formula
Slovin’s Formula is used only if the study is to estimate a
population proportion.
Read the article “On the Misuse of Slovin’s Formula”.
8. Slovin’s Formula: Sample computation
Assume: N=3000 and e = 5% or .05
• For a population of N=3000, you need a
sample size of n=353.
• Many researches including theses and
dissertation use the so-called Slovin’s
Formula.
• See table at the right and observe the
values of the sample size (n) as population
size (N) increases.
N n
3,000 352.94
5,000 370.37
100,000 398.41
500,000 399.68
1,000,000 399.84
5,000,000 399.97
100,000,000 400.00
1,000,000,000 ?
1,000,000,000,000 ?
1,000,000,000,000,000 ?
9. Slovin’s Formula: How the formula was derived?
To make inferences on the POPULATION
PROPORTION (P) under SRSWR, Cochran(1977)
presents the following formula for sample size:
where:
N = population size
z = the standard normal variate based on
confidence level
e = margin of error
p = estimate of population proportion (P)
10. To arrive at Slovin’s Formula:
• In a standard normal distribution, z = ± 1.96 (or z = ±2.0) at α =.05.
• In the absence of any prior knowledge about P, the conservative approach is
to maximize the p(1-p) of the equation:
p p(1-p)
0.10 0.09
0.20 0.16
0.30 0.21
0.40 0.24
0.50 0.25
0.60 0.24
0.70 0.21
0.80 0.16
0.90 0.09
1.00 0.00
The so-called Slovin’s Formula!
• Substitute z= ± 2.0 and p = .5, yields:
• From the table, p(1-p) is maximum at p=.5 .
11. Misuse of Slovin’s Formula
Note that Slovin’s Formula is appropriate only in
determining sample size when the research is to estimate a
population proportion using 95% confidence coefficient
and the population proportion to be estimated is
suspected to be close to .5.
Also, note that it was Yamane(1967) who devised the
formula. Seemingly, no Slovin in the history of Statistics.
12. II. Sample Size Issues
b. Power Analysis
Conduct a power analysis if the research will use statistical
tests.
Power analysis is finding the optimal combination of:
• Sample size,
• Effect Size,
• tolerable errors in decision (Type 1 & Type II Errors), and
• complexity of the statistical test.
NOTE: Each Statistical Test has its own formula, but complex;
hence, we need to use a power analysis software.
13. II. Sample Size Issues
c. Software for Power Analysis
FREE: GPower
Some Commercial
Softwares
14. III. Statistical Modeling Issues
a. Statistical Models were created with underlying
Assumptions/Conditions
Moderate violations of the assumptions (e.g., normality) have
little or no effect on substantive conclusions in most instances
(Cohen, 1969).
But, do not ignore the assumptions/conditions. Check them!
Also, handle with care the other factors (multicollinearity,
missing values, and Outliers/influential points). Your results
might be distorted by these.
15. IV. Updates of Statistical Tools
• We always aim for excellence, but do we update our
Syllabus?
Sample Old Statistical Models (most don’t know but pretend
to be knowledgeable; others don’t care; ):
Path Analysis – Developed by Sewall Wright (1918, 1920)
Confirmatory Factor Analysis (CFA) – Developed by Joreskog
(1969)
Structural Equation Modeling – early development was due
to Joreskog (1969, 1973), Keesling (1972), Wiley (1973)
16. V. Teaching Statistics in the Graduate
School: MANUAL COMPUTATION
Non Math students are forced to
manipulate complex mathematical
formulas.
Are these things difficult,
tedious/boring/dull/tiresome to
majority of them?
17. V. Teaching Statistics in the Graduate
School: Manual Computation
Students need the
Statistical Tools to prepare
them to do research.
But why ask them to do
the Math?
18. V. Teaching Statistics in the Graduate
School: Manual Computation
If non-Math People
are forced to do complex
math computation, their
attitude about the
subject becomes
negative.
Consequence #1:
19. V. Teaching Statistics in the Graduate
School: Manual Computation
After graduation( when he
returns to his institution), he will
handle real projects with large sample
sizes ( say, n>1000 or n> 5,000).
In school, a
graduate student is
trained to perform data
analysis manually with
small samples (e.g., n=30,
50, or <100).
20. V. Teaching Statistics in the Graduate
School: Manual Computation
After the comprehensive
exam, the graduate student
hired a statistician to do the
data analysis for his/her
thesis/dissertation.
What happen?
Graduate Student
passed the
comprehensive exam
in statistics.
21. V. Teaching Statistics in the Graduate
School: Manual Computation
Only few topics may be tackled because up to
80% of the time were wasted to do the Math.
No more time to discuss
for the advanced statistical models
Consequence #4:
22. V. Teaching Statistics in the Graduate
School: Manual Computation
QUESTION! Can manual
computation help boost the
research capability and productivity
of graduate students and faculty?
23. Why not Replace the Math with
statistical software.
Why not focus the discussion on the
following:
Concepts and Applications: (e.g. when
to use the statistical test)
How to carry out the computation
using a software?
How to present and interpret the
results?
VI. Use of Statistical Software
in Teaching Statistics
24. Through the use of statistical software,
Students’ Math anxiety is no longer an issue;
Students have more time to do the right way of data
analysis(from cleaning, exploratory data analysis, …,
to checking of assumptions)
Students have more time to practice conducting
statistical data analysis using real data;
Students have more time to practice in writing the
results;
More topics can be discussed in class.
25. Thank you for Listening!
My contact details:
Prof. Johnny T. Amora
De La Salle-College of Saint Benilde
Taft Avenue, Manila, Philippines
Email: johnny.amora@gmail.com