[06.05] Sample Size Estimation V2.pdf

Determining the minimum requirement for the number of samples in a research

What is sample size calculation?

A problem

What is the consequence of having fewer samples than the minimum requirement?

Acceptable, as this increases power

How is exceeding the sample size requirement viewed?

Validity, reliability, and ethical conduct of health research

What three aspects of health research does sample size calculation directly impact?

Precision and Reliability, Statistical Power, Cost-efficiency, and Ethical Considerations

What are the four main areas of importance for sample size calculation?

Minimizing random variation in the data

How does sample size calculation ensure that study results are precise and reliable?

By having sufficient samples

What action ensures precise and reliable results by minimizing random variation?

Random error

What type of error is present in samples, suggesting a probability of missing the true parameter value with fewer samples?

Results approach the true parameter value

What benefit does a greater sample size ensure regarding study results?

Detect a true effect (or significant association/difference)

What ability of the study does statistical power determine?

Reducing the risk of false-negative findings

What is the consequence of determining the ability to detect a true effect (statistical power)?

Successfully reject a false null hypothesis

When is power achieved?

Type II error

What is the failure to reject a false null hypothesis (a false-negative) called?

Less power

What is the consequence of having fewer samples?

Committing a Type II error

What is the consequence of having an underpowered test due to insufficient samples?

An underpowered test

What must researchers rule out as the possible reason for failing to reject the null hypothesis when a true null hypothesis is present (i.e., no statistically significant difference)?

Sufficient sample size

What is necessary to rule out an underpowered test?

Optimize resource utilization

What process does sample size calculation help with regarding cost-efficiency?

Minimum sample size required to achieve study objectives

What requirement does sample size calculation determine to optimize resources?

To set your budget

What can knowing the minimum sample size requirement allow you to do?

Minimizes unnecessary exposure of participants to potential risks

How does sample size calculation relate to ethical considerations?

Avoiding overly large or small sample sizes

What specific practice regarding sample size minimizes unnecessary exposure of participants to potential risks?

Experimental studies (e.g., vaccine efficiency, medicines, etc.)

What type of studies make minimizing participant exposure especially true?

To avoid disrupting operations

In observational studies (e.g., NCDs), why is contacting the entire population in the hospital unnecessary?

Population vs. Sample, Variability (Standard deviation), Effect size, Confidence Level, and Power

What are the five Key Concepts listed for sample size calculation?

Measure of dispersion

What is Variability (Standard deviation)?

Reflects the degree of spread or dispersion of values within a set of data

What does variability reflect?

Larger sample size

What is required to achieve the same level of precision if variability is higher?

Smaller sample size

What is required to achieve the same level of precision if variability is lower?

Older literature

What must sample size calculation reference to assume parameters like mean, percentage, and variability?

Magnitude of the difference or relationship between groups and variables

What is Effect size?

Smaller sample size

What is required if the effect size is larger?

The distance between the percentages displayed by both groups

What would the effect size be when comparing the prevalence of hypertension between smokers and non-smokers?

High

If the difference in hypertension prevalence between smokers (70%) and non-smokers (20%) is 50%, how is the effect size described?

A sample size of 30,000

If the effect size is 1% (40% vs. 39% prevalence of hypertension), what sample size should be expected?

Odds ratio

What is another measure of effect size mentioned?

Smaller sample size

What is implied if the odds ratio is more than 5 regarding the required sample size?

To lessen the required sample size

Why are the indicators (confidence level, power) sometimes manipulated?

Probability that the interval estimate for a population parameter from a sample will contain the true population parameter

What is the definition of Confidence Level?

95%

What is the usual setting for the confidence level?

It changes the Type I error (alpha error)

What is the consequence of changing the confidence level?

Significance level

What is the complement of the confidence level?

Probability of committing Type I error or alpha error

What is the significance level?

0.05 or 5%

What is the usual setting for the significance level?

Probability of correctly rejecting the false null hypothesis

What is Power?

Ability to find true effect or difference between the factor and outcome if it exists in the population

What does power indicate?

0.80 or 80%

What is the usual setting for power?

Type II error or beta error

What is the complement of power?

Single proportion, Comparing 2 proportions (Unmatched case-control, Cross-sectional/cohort/RCT), and Comparing 2 means

What are the three main sample size calculations focused on in the lecture?

If the outcome of interest is quantitative

When is the 'Comparing 2 means' calculation used?

Open Epi

What is the free online tool that can be used to perform statistical tests and sample size calculation?

Slovin’s formula

What is the simplistic version of the sample size calculation for a proportion of % frequency in a population?

0.5 or 50%

What estimated proportion ($\hat{p}$) does Slovin’s formula assume?

Maximum sample size

What results from assuming an estimated proportion of 0.5 or 50% in Slovin’s formula?

Smaller sample size

What results from veering away from 0.5 (higher or lower) in the estimated proportion?

95% confidence level and 1.96 Z-score

What two values are used in the sample size calculation for a proportion of % frequency in a population?

5% error/level of significance

What value is used for the level of significance in the sample size calculation for a proportion of % frequency in a population?

Sample Size ($\text{n}$)

What variable is being computed for in sample size calculation?

Design Effect ($\text{deff}$)

What variable stays at 1.0 in simple random sampling?

To account for variability in complex random surveys

Why is the Design Effect increased?

Cluster sampling

When must the Design Effect be increased (i.e., when using this sampling type)?

Arbitrary and often based on assumptions on the parameters

What are the increase in design effect based on?

Around 1000

When must Population Size ($\text{N}$) be computed?

1 million

What value should N be kept at in populations estimated to be over 1 million (e.g., Pasig City)?

From literature or close to literature

Where must the Estimated Proportion ($\hat{p}$) be taken from?

Categorical variable (e.g., prevalence of disease or level of awareness)

What type of variable is used when applying the formula for single proportion?

$\text{q} = 1 - \hat{p}$

What is the formula for $\text{q}$?

Desired absolute precision or absolute level of precision ($\text{d}$)

What parameter is set by the researcher and can be adjusted for logistical constraints?

A range of possible values from 45% to 55%

If the absolute precision is set to 5% with a proportion of 50%, what is the range of possible values?

Statisticians prefer to stay within 5%

What is the preferred absolute precision by statisticians?

To see if the sample size is decreasing

Why might a researcher adjust the absolute precision to numbers like 6% or 7%?

If estimates are too high

When might the researcher adjust the absolute precision to handle logistical constraints?

One million

What is the default value on OpenEpi for population size?

When the population size is small

When does computing the population size reduce the sample size?

An educated guess of the % of the population with the outcome of interest

What is the Anticipated % frequency ($\text{p}$)?

50%

What percentage results in the largest sample size if the researcher is unsure of the percentage?

How wide (in absolute terms) you would like the confidence interval to be around your point estimate

What are the Confidence Limits as +/-percent of 100?

1

What is the Design Effect (DEFF) if simple random sampling is used?

Frequently > 1 (ranging 2~10)

What is the typical range for DEFF if a cluster-type survey is used?

384

What is the resulting sample size if the prevalence is 50% and absolute precision is 5% in the example?

200

What is the minimum acceptable sample size that Prof. Sto. Niño is happy with for ASMPH students?

Manipulate the confidence limits or the absolute precision

What is the suggested way for ASMPH students to reach the desired sample size if 384 is too high?

Type II error

What type of error is increased when statistical power is decreased?

False

Is it True or False that in sample size calculation for a single proportion, the deff is increased when employing simple random sampling?

Desired absolute precision or absolute level of precision

What is the parameter that the researchers themselves can adjust for logical constraints?

Unmatched Case-Control Study

What study design compares two proportions based on the disease of interest?

Case (disease of interest) and Control (healthy population)

What are the two groups compared in an unmatched case-control study?

Randomly selected

How is the control group selected in an unmatched case-control study?

Proportional exposure between a case and control

What can be measured in a case-control study?

Do not touch; stays at 95 only

What is the rule of thumb for the Confidence Level in comparing two proportions?

80%

What is the lowest allowable power for an unmatched case-control study?

1:1 ratio

What is the value of the Ratio of Controls to Cases if they are equal?

Precision and more generalizability among the controls

What is the advantage of increasing the ratio of controls to cases (e.g., 4:1)?

Odds Ratio, Percent of Controls Exposed, and Percent of Cases with Exposure

What three key values are needed for Unmatched Case-Control Sample Size calculation (at least two needed for OpenEpi computation)?

Hypothetical Proportion of Controls with Exposure

In a study on smoking and cancer, where is the proportion of smokers who are free from cancer inputted?

Hypothetical Proportion of Cases with Exposure

In a study on smoking and cancer, where is the proportion of smokers among breast cancer patients inputted?

The likelihood of cancer among those who smoke is twice compared to those who do not smoke

What is the interpretation if the Odds Ratio is 2 (e.g., smoking and cancer)?

Kelsey, Fleiss, and Fleiss with CC (Corrections)

What are the three resulting sample size values shown in OpenEpi for comparing 2 proportions?

Whichever is the highest

What is the rule of thumb among the three resulting sample size values (Kelsey, Fleiss, and Fleiss with CC)?

Look for studies with a larger effect size

What adjustment should be made if the total sample size (e.g., 288) is too high (besides not changing alpha/power)?