Method of clinical medical statistic (3)

Medical School Pathology, Chapter 2a

3. The estimation of example content

The estimation principle of example content is to show in research conclusion has certain reliability (examine efficiency) the affirmatory and least example example number on the foundation. Estimate example content, the purpose is to be below the premise that assures proper definition, decide least observation unit is counted.

In clinical trial research, no matter be experimental group,still contrast group need to have suffer a certain quantity ofly try an object. Because,this is same processing is in kind of experiment to suffer differently trying the experimental effect that shows on object body is to existing to mutate. Only by experiment observation result or individual the experimental effect that suffers expression of place of the person that try to come out cannot show what issue. Must pass a certain quantity of repetition observation ability shows the objective law sex with overall and real research come, and can be made to error of sampling estimate objectively. Say to repeat observation time commonly more, error of sampling is smaller, the reliability of observation result is higher. A certain quantity of repetition still can have the effect that the part offsets jumbly element is affected, those who enhance across block can compare a gender. But repeat observation time more (namely example content is bigger) the labor power that experiment place should consume, material resources, financial capacity and time are jumped over much, may make the experiment considers to become impossible. And, example content still can increase the difficulty that controls experiment observation condition greatly too, introduce likely be not random error, bring to observation result slant quality (Bias) . A when repeat a principle in fact of experimental design decline so serious problem is how content of scientific and reasonable affirmatory example. Because be when number of each parallel form example is equal,undertake statistic concludes efficiency is highest, because this is below most circumstance,press content of each groups of example to estimate equally. But fall in individual situation, ask content of each groups of examples estimates by certain proportion possibly also.

Determine the premise condition of example content:

1) examine certainly horizontal α

Make the first kind of wrong probability certainly, namely significance level, take α =0.05 commonly, still should be being made clear at the same time is odd side examine or be bilateral examine, here α is smaller, reckon example content is bigger.

2) examine certainly efficiency (1, β )

β is the probability that makes the 2nd kind of mistake, the requirement examines efficiency is bigger, what want example content bigger also, take β =0.10 commonly, examine efficiency 1, β =1-0.10=0.90, when clinical research is designed, examine efficiency is unfavorable under 0.75, , if under 0.75, possible research cannot reflect an overall true difference as a result, the negative result with likelihood occurrence actual blame.

Medical is statistical in software, use PASS software mostly to the estimation of example content, current version is: NCSS 2004 And PASS 2005v2.0.0.462 (download address Http://www.9iv.com/down/soft/1184.htm) . This software is compositive NCSS and PASS two component.

The example content that is based on dependency coefficient analysis estimates a method, if the graph chooses only group dependency,analyse menu.

Inside the window of dependency analysis command in PASS " DATA " label page undertakes configuring parameter:

Find (SolveFor) : Can choose to examine here horizontal α , hold degree of β , the output result such as correlation coefficient R and sample content N;

R0 (BaselineCorrelation) :

R1 (Alternative Correlation) : Example examines beforehand dependency

Alpha (Significance Level) : Examine horizontal α

Beta (1-Power) : Examine efficiency, for 1, β

After parameter of make choice of, click move (computation) pushbutton, computation will be in as a result " NCSS OUTPUT " show in the window, as follows (here chooses R1=0.5) :

One CorrelationPower Analysis

Page/Date/Time1
2009-2-2712:30:35

Numeric Resultswhen Ha: R0<>R1

Power
NAlphaBeta
R0
R1

0.81394290.050000.186060.000000.50000

References

Graybill, franklin.1961. An Introduction To Linear Statistical Models. McGraw-Hill.New York, new York.

Guenther, william C.1977. ‘Desk Calculation Of Probabilities For The Distribution Ofthe Sample Correlation

Coefficient‘ , theAmerican Statistician, volume 31, number 1, pages 45-48.

Zar, jerrold H. 1984.Biostatistical Analysis. Second Edition. Prentice-Hall. EnglewoodCliffs, new Jersey.

ReportDefinitions

Power Is Theprobability Of Rejecting A False Null Hypothesis. It Should Beclose To One.

N Is The Size Of Thesample Drawn From The Population. To Conserve Resources, it Shouldbe Small.

Alpha Is Theprobability Of Rejecting A True Null Hypothesis. It Should Besmall.

Beta Is Theprobability Of Accepting A False Null Hypothesis. It Should Besmall.

R0 Is The Value Ofthe Population Correlation Under The Null Hypothesis.

R1 Is The Value Ofthe Population Correlation Under The Alternativehypothesis.

SummaryStatements

A Sample Size Of 29achieves 81% Power To Detect A Difference Of -0.50000 Between Thenull

Hypothesiscorrelation Of 0.00000 And The Alternative Hypothesis Correlationof 0.50000 Using A

Two-sided Hypothesistest With A Significance Level Of 0.05000.

ChartSection