27 November 2008

Sample Sizes and Power - Besar Sampel dan Kekuatan

Dalam pengambilan sampel untuk sebuah penelitian, besarnya sampel sangatlah berpengaruh dan memberikan sumbangan yang tidak sedikit terhadap kualitas sebuah penelitian. Bagaimana hal tersebut bisa terjadi? Bagaimana menentukan besar sampel yang sesuai dengan tujuan dan desain penelitian kita? Software apa yang bisa membantu kita? Apa saja kesalahan-kesalahan yang kadang kita buat dalam menentukan besar sample tersebut?


Berikut ini adalah peta yang saya buat dari apa yang saya pelajari mengenai Sample Size and Power dari presentasi Prof. dr. Mohammad Hakimi, SpOG, PhD.

Power point mengenai Besar Sampel dan Kekuatan dapat didownload disini (informasi mengenai software sample size ada di dalamnya)


Oke Semoga Bermanfaat

Wassalam

Salam Sukses


Ini adalah format ounlinennya:

SSAMPLE SIZE (SS) AND POWER

I. SS Planning

A. forces

1. specification for outcome variable

2. clinically meaningful effect size

3. planned stat procedure

B. leads to a specific recruitment goal

C.

1. encourages

a) development of appropriate

timelines

budgets

2. discourage

a) performance of small & inconclusive studies

D. to estimate SS requirement

1. specify hypothesis

a) null

b) alternative

2. select appropriate stat test

3. determine min effect size

4. estimate

a) SD (for continuous outcome)

b) baseline risk/incidence/ prevalence (for dichotomous)

5. set limit error

a) type I (alpha)

b) type II (beta)

II. Hypothesis Testing

A. the decision matrix

1. the confidence level (1-alpha)

= we accept null, reality null true

2. type I error (alpha)

= we reject null, reality null true

3. type II error (beta)

=we accept null, in reality null false

4. power (1-beta)

=we reject null, in reality null false

III. Software

A. epiinfo

- for cohort and unmatced case control

- www.cdc.gov/epiinfo/

B. southwest oncology group

- for clinical trials

- www.swogstat.org/statoolsout.html

C. UCLA

- for Poisson analysis, correlation coefficients, Fisher's exact test, and more

http://calculators.stat.ucla.edu/powercalc/

D. Commercial products

- http://www.power-analysis.com/

- http://www.statsol.ie/demos.htm (NQuery Advisor)

- http://www.ncss.com/pass2002upgrade.html

E. Commonly Used

1. Power and Sample Size Calculation

a) ..\..\Power and Sample Size\pssetup.exe

IV. What To do when SS is fixed

= SS may determined before study was planned

A. work backward from the SS

1. Estimate the effect size that can be detected at a given power (usually 80%)

2. estimate the power to detect a given effect

=less commonly

V. Strategy for Minimizing SS and Maximizing Power

A. use continuous variable

B. Use paired measurements

C. Use more precise variables

D. Use unequal group sizes

E. Use a more common outcome

VI. how to estimate SS when there is insuff info

A. Extensive search for previous findings

B. Doing a small pilot study

C. Dichotomize a continuous variable when mean & SD are in doubt

VII. How to Get By with a Smaller Sample

A. double check

1. your assumptions:

a) type I error rate

b) type II error rate

2. your hypothesis

a) one tailed?

b) two tailed?

3. your desired effect size

a) Is the intervention have an effect of this size?

b) Can I settle for finding a larger effect?

B. Use a more frequent dichotomous outcome variable

1. Less serious outcome (don=t sacrifice clinical face validity)

2. Composite outcome (watch for undesired heterogeneity)

3. Lengthen follow-up period

4. Select higher-risk study subjects

C. To decrease variability (increase reliability) of continuous outcome measures, use:

1. A continuous variable

2. More precise, reliable equipment or data collection techniques

3. Paired (pre-post or matched) measurements to reduce variability

D. Increase your sample size cheaply:

= Consider unequal group sizes (2-3 control per case) to increase efficiency in a case control study

E. Compromise or get more money

VIII. Common MIstakes

A. Failure to discuss sample size

B. Unrealistic assumptions

C. Failure to explore SS requirements

D. Treating Type I (<0.05)>

E. Failure to account for attrition

1 komentar:

PS 18 Februari 2009 11.48  

assalamu'alaikum salam kenal mba. cita - cita saya juga pengen jd peneliti dan dosen, skrg br ngajar 1 mt kuliah di stikes, gmn mba ada saran? email saya sari_ansa@yahoo.com, sariansa@blogspot.com. saya br mau selesai S2 di mutu yankes FKM UI

AddThis

Share |

  © Free Blogger Templates 'Greenery' by Ourblogtemplates.com 2008

Back to TOP