Elementary Medical Statistics Second EditionPDF电子书下载

CHAPTER Ⅰ QUESTIONS FOR USE IN PLANNING INVESTIGATIONS AND IN EVALUATING REPORTS 1

The Purpose of the Questions 1

Statistics in Medicine 2

Abuse and Misuse of Statistics 3

Statistics and Mathematics 5

The Research Worker and Statistical Techniques 5

Sources of Variation 7

The Ten Groups of Questions 8

Ⅰ．The Nature of the Research and the Attitudes of the Researchers．QⅠ-1．Is our project actually research，or is it a demonstration or an educational program or provision of improved services？ 8

QⅠ-2．Are we going to turn a blind eye on some awkward bit of evidence？ 9

QⅠ-3．Will we be hindered by our desire for a simple answer？ 9

QⅠ-4．If our research will entail experimentation on human beings，what will be its ethical justification？ 9

QⅠ-5．If we intend to conduct experiments on human beings how are we to insure that ethical conflicts will not vitiate our results？ 13

CHAPTER Ⅱ PURPOSE AND GENERAL METHOD OF INVESTIGATION 15

QⅡ-1．What is the immediate purpose of our investigation？What are the specific questions to be answered by it？ 15

QⅡ-2．What is the more remote(more general or ultimate)purpose of our investigation？ 15

QⅡ-3．Will the variables that we observe be the variables that we really wish to know about？ 17

QⅡ-4．Will the investigation be an experiment in the strict sense，or a survey，or a mixture of the two？ 19

QⅡ-5．If the investigation is a survey，will it be anecdotal or evaluational？ 23

QⅡ-6．If the investigation is a survey，will it be retrospective，prospective or instantaneous(cross-sectional)？ 24

QⅡ-7．If the investigation is a survey，what is the imaginary experiment which，if it were possible，we would，perform instead of the survey？ 26

QⅡ-8．What will be our contrast-groups？How comparable will they be initially？ 26

QⅡ-9．Are we trying to answer too many questions by one piece of research？ 27

QⅡ-10．Do we intend to conduct a pilot study before starting our definitive investigation？ 28

QⅡ-11．Have the questions that our research is intended to answer been already answered by someone else？ 28

QⅡ-12．Is this research outside our accustomed field？ 28

CHAPTER Ⅲ THE POPULATION AND SAMPLING 29

QⅢ-1．What is the initial definition of our population-the kind of subject or material which is to be studied，and tc which our conclusions will be applicable？ 29

QⅢ-2．Is the initial definition of the population adequate for the selection of subjects？ 29

QⅢ-3．What other populations are excluded by the definition of our population？ 30

QⅢ-4．Will the definition of the population be observed strictly throughout the study？ 31

QⅢ-5．How may the study itself affect the definition of the population？ 31

QⅢ-6．By what technique will we obtain our samples？ 32

QⅢ-7．What is the danger of haphazard selection？ 34

QⅢ-8．What is the danger of a purely systematic selection method？ 35

QⅢ-9．If our samples are not strictly random，how can we generalize，i．e．，argue from them to their populations？ 36

CHAPTER Ⅳ SUBDIVISION OF POPULATION AND SAMPLES 39

QⅣ-1．Do we intend to subdivide our original population in order(a)to study a sample from each class separately，or(b)to confine our study to one class？If so，why？ 39

QⅣ-2．Apart from the factor under test，what are the variables that are associated with differences in the phenomena to be observed？ 42

QⅣ-3．Which of the variables listed in answer to Question Ⅳ-2 will we use as a basis for subdivision of the population？Why not the other variables？ 44

QⅣ-4．If we choose a measured variable(e．g．，age)as a basis for classification，how will we decide on the coarseness or fineness of the subdivision？ 44

QⅣ-5．If our research is an experiment in the strict sense，are we going to make paired comparisons within the same subject or between matched subjects？If so，why？ 48

QⅣ-6．What are the risks of paired comparisons in an experiment？ 49

QⅣ-7．If our research is a survey，what will be the advantages and risks of a matched comparison？ 51

QⅣ-8．If our research is an experiment，do we wish to increase its efficiency by utilizing a rather more complicated design than the two-group or paired comparison？If so，what are the advantages and risks？ 51

QⅣ-9．What two effects of variation should we remember at every point in an investigation？ 53

CHAPTER Ⅴ SKELETON LAYOUTS FOR RESULTS 54

QⅤ-1．What skeleton layout must we prepare to receive our results？ 54

QⅤ-2．Are measurements to be presented and analyzed as measurement data or as frequency data？What will determine the choice？ 58

QⅤ-3．If we intend to present measurements as frequency data，when will we select the dividing lines between the classes？ 59

QⅤ-4．Will our sampling units be independent of each other，except insofar as the randomization brings them together？ 59

QⅤ-5．Are we going to mix the sampling units？ 61

QⅤ-6．Will the denominators be correct？ 63

QⅤ-7．What will be the groups exposed to risk？ 64

QⅤ-8．Will the difference between＂incidence＂and＂prevalence＂mislead us？ 67

QⅤ-9．What do we expect to be the order of magnitude of the frequencies and measurements that we observe，and of quantities that we calculate during the analysis？ 69

CHAPTER Ⅵ INTERPRETATION AFTER AN EXPERIMENT 70

QⅥ-1．What exactly，is the hypothesis that we wish to test by our experiment？ 71

QⅥ-2．By what rules are we going to insure that the causal interpretation after the experiment will take the form：＂Either the randomization or the factor under test＂？ 72

QⅥ-3．Is randomization necessary if we are not going to apply a statistical test？ 72

QⅥ-4．What is the risk in trusting to our preexisting knowledge of variability and to our way of conducting the experiment，instead of using randomization？ 73

QⅥ-5．What are the risks in using the method of＂alternates＂instead of true randomization？ 75

QⅥ-6．What instrument are we going to use for randomization？ 78

QⅥ-7．What sources of variation must we remember in planning the randomization？ 82

QⅥ-8．What are we going to randomize，and how？ 82

QⅥ-9．Are we going to examine our treatment assignments in order to see if the randomization has＂worked well＂？ 86

QⅥ-10．How are we going to insure that the randomization will be solely responsible for bias throughout the experiment？ 87

QⅥ-11．If we are trying to find whether there is a difference in the effects of two or more treatments of any kind，what are the two types of error that we may make in drawing our conclusions？ 88

QⅥ-12．If we intend to apply a test of＂statistical significance＂to our data，how will we know that it is a legitimate test？ 90

QⅥ-13．If a legitimate test is applied to our data and gives a verdict＂statistically significant difference，＂what will it tell us？ 91

QⅥ-14．If a legitimate test is applied to our data and gives a verdict＂nonsignificant difference，＂what will it tell us？ 93

QⅥ-15．If we are going to apply a test of statistical significance，what level of significance will we choose？ 93

QⅥ-16．If the verdict of a test is＂significant＂at our predetermined level，what will we do in consequence thereof？ 94

QⅥ-17．If the verdict of a test is＂not significant＂at our predetermined level，what will we do in consequence thereof？ 97

QⅥ-18．What will make us decide to terminate the experiment？ 98

QⅥ-19．Are we going to make more than one comparison in the same data？If so，are significance tests legitimate？ 102

QⅥ-20．If we are comparing more than two treatments，how may significance tests mislead us？ 103

QⅥ-21．How will we estimate what our results really tell us，numerically，about the population represented by our sample？ 106

QⅥ-22．Is a＂null hypothesis＂an appropriate concept for our experiment？ 109

CHAPTER Ⅶ INTERPRETATION AFTER A SURVEY 111

QⅦ-1．What is the most obvious risk in a cause-and-effect interpretation after a survey？ 111

QⅦ-2．What is the essential difference，with regard to causal interpretation，between a survey and an experiment in the strict sense？ 112

QⅦ-3．Are we going to apply a statistical significance test after our survey？If so，what will it tell us？ 113

QⅦ-4．What will be the logic of our inference after a survey？ 113

QⅦ-5．How may differences in admission rates(selection rates)produce spurious results in a survey？ 117

QⅦ-6．What are we going to do about the risk of bias due to interplay of admission rates in our survey？ 124

QⅦ-7．Under what conditions can we omit a contrast-group in a survey？ 125

QⅦ-8．Are we going to use a previous series of cases in evaluating a new treatment？ 126

QⅦ-9．Are we going to use patients as a contrast-group in a survey？ 127

QⅦ-10．Are we going to use official vital statistics in connection with our study？If so，what precautions should we take？ 128

QⅦ-11．If，instead of taking a particular hypothesis to clinical records，we search among them for possible relationships，are there any fruitful methods of procedure？ 132

QⅦ-12．In view of the uncertainties of research by survey methods，ought we to avoid these methods as much as possible？ 133

CHAPTER Ⅷ SAMPLE SIZES 134

QⅧ-1．How large should our sample be？ 134

QⅧ-2．Having decided on the difference to be detected，and on our Type Ⅰ and Type Ⅱ errors，how are we to find our required sample sizes for a binomial two-treatment experiment or survey？ 135

QⅧ-3．If our data are to be measurement data how will we estimate required sample sizes？ 136

QⅧ-4．If we wish to use a sequential design how are we to estimate the required sample size？ 136

QⅧ-5．If we have to use unequal samples is there risk of bias？ 136

QⅧ-6．Do small samples introduce greater risk of biases than large samples？ 137

QⅧ-7．Will our available subjects and other facilities enable us to obtain a sample that is large enough for our purpose？ 138

CHAPTER Ⅸ COLLECTING，RECORDING AND EXAMINING THE DATA 139

QⅨ-1．Have we overlooked any of the influences that observers and other personnel may have on our investigation？ 139

QⅨ-2．If more than one observer will be involved，how may their differences affect the results？ 140

QⅨ-3．If we are conducting an animal experiment how may the condition of the animals affect our observations？ 141

QⅨ-4．How may unidirectional changes affect our observations？ 141

QⅨ-5．How may fluctuations affect our observations？ 142

QⅨ-6．How will we correct for unidirectional and fluctuating changes？ 143

QⅨ-7．Are we going to deceive ourselves by starting to time a phenomenon at the wrong point？ 144

QⅨ-8．How will we correct for differences in length of exposure to risk？ 145

QⅨ-9．Are we going to use routine clinical records in our investigation？ 146

QⅨ-10．Are we going to use questionnaires in our investigation？If so，what problems will arise？ 147

QⅨ-11．What are some of the precautions that we should take in planning data sheets？ 149

QⅨ-12．How will we try to prevent defects in observing and recording？ 151

QⅨ-13．How will we minimize misstatements in recording？ 151

QⅨ-14．Will our observational methods be sufficiently accurate for our purpose？ 153

QⅨ-15．If we are going to estimate observational variation will our estimate include all its components？ 160

QⅨ-16．How will we know that an event reported in a clinical experiment has really happened？ 160

QⅨ-17．Are we aiming at too great precision？ 161

QⅨ-18．How will we insure，for all sampling units，an equal exposure to observation？ 162

QⅨ-19．Will we be consistent regarding precision in the various parts of the investigation？ 162

QⅨ-20．Are we going to be deceived by the structure of our measurements？ 163

QⅨ-21．How，in general terms，are we going to handle our data during analysis？ 168

QⅨ-22．What are we going to do about＂significant figures＂and decimal places？ 168

QⅨ-23．What guidelines should we follow in preparing tables and graphs？ 170

QⅨ-24．Are we going to be deceived by average i trends？ 171

QⅨ-25．Are we going to confuse absolute and relative rates of change？ 173

QⅨ-26．Are we going to be deceived by the＂goodness of fit＂of our data to a mathematical formula？ 173

QⅨ-27．Are we going to reject observations that seem aberrant？If so，on what basis？ 175

CHAPTER Ⅹ LOST INFORMATION 178

QⅩ-1．Will large numbers compensate for lost information？ 178

QⅩ-2．In a controlled trial what can be done about refusals to participate？ 179

QⅩ-3．In a survey is it safe to make up for refusals and losses by enrolling more subjects？ 180

QⅩ-4．What should be our maximum percentage of losses？ 180

QⅩ-5．How will we try to minimize losses？ 181

QⅩ-6．Will we be able to retrieve bits of missing information？ 182

QⅩ-7．How will we deal with the problem of lost information when analyzing our results？ 182

QⅩ-8．What are we to do about lost information in an experiment on nonhuman material？ 186

CHAPTER Ⅺ ESTIMATING POPULATION PERCENTAGES FROM SAMPLES OF FREQUENCY DATA-AN INTRODUCTION TO RANDOM PROCESSES 187

CHAPTER Ⅻ COMPARISON OF SAMPLES OF FREQUENCY DATA 218

CHAPTER ⅩⅢ VARIATION BETWEEN MEASUREMENTS 241

CHAPTER ⅩⅣ COMPARISONS BY MEASUREMENT 270

CHAPTER ⅩⅤ CONCOMITANT VARIATION AND TRENDS 308

CHAPTER ⅩⅥ SUPPLEMENTARY NOTES ON TECHNIQUES AND FURTHER STUDY 339

REFERENCES 353

APPENDIX 357

Table Ⅰ．Limits of Binomial Population Percentages of X＇s Estimated from Random Samples-No．of X＇s in Sample：0-50；Sample Sizes：1-100 358

Table Ⅱ．Limits of Binomial Population Percentages of X＇s Estimated from Random Samples-Percentage of X＇s in Sample：50-1，Sample Sizes：100-1000 362

Table Ⅲ．Minimum Contrasts Required in Fourfold Contingency Tables to Insure a Type Ⅰ Error of Not More than 5 Per Cent(2．5 Per Cent in Each Tail) 364

Table Ⅳ．Minimum Contrasts Required in Fourfold Contingency Tables to Insure a Type Ⅰ Error of Not More Than 1 Per Cent(0．5 Per Cent in Each Tail) 366

Table Ⅴ．Percentages of Successful Experiments(%S)in Relation to Sample Sizes and to Percentages of X＇s in Populations A and B 368

Table Ⅵ．Signed-ranks Test(Wilcoxon) 371

Table Ⅶ．The Gaussian Distribution-Frequency of Random Samples Lying Beyond Specified Multiples of the Standard Deviation Measured Above and Below the Mean 372

Table Ⅷ．Two-sample Ranks Test(Wilcoxon-White) 373

INDEX 375