1 copyright © 2011 by saunders, an imprint of elsevier inc. chapter 11 understanding statistics in...
TRANSCRIPT
1Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Chapter 11
Understanding Statistics in Research
2Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Clinical Uses of Statistics
Reading or critiquing published research Examining outcomes of nursing practice by
analyzing data collected in clinical site Developing administrative reports with
support data Analyzing research done by nursing staff and
other health professionals at a clinical site Demonstrating a problem or need and
conducting a study
3Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Stages in Data Analysis
1. Prepare data for analysis.2. Describe the sample.3. Test the reliability of measurement methods.4. Conduct exploratory analysis.5. Conduct confirmatory analysis guided by
hypotheses, questions, or objectives.6. Conduct post hoc analyses.
4Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Preparing the Data for Analysis
1. Enter data into the computer using means designed to reduce errors.
2. Clean the data to ensure accuracy.3. Correct all identified errors.4. Identify missing data points.5. Add missing data when possible.
5Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Describing the Sample
Purpose: to obtain as complete a picture of the sample as possible Determine frequencies of variables related to
sample• Age• Education• Gender• Health status• Ethnicity
6Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Describing the Sample (cont’d)
Examine averages and variation of demographic variables.
If there are study groups, compare using variables such as age, education, health status, gender, and ethnicity.
Determine the comparability of groups. If groups are not comparable, planned
comparative analyses cannot be performed.
7Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Testing Reliability of Measurement
Examine reliability of study scales before testing hypotheses, questions, or objectives using Cronbach’s alpha coefficient.
Values should be at least 0.70.
8Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Cronbach’s Alpha Coefficient
Tests internal consistency of measurement scale
To what extent is the measure a true reflection of subject’s responses?
Reliability of 0.7 is the lowest acceptable alpha.
This means that 70% of the time you can trust the score to accurately reflect what is being measured.
9Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Conducting Exploratory Analysis
Determine the nature of data in variables used to test hypotheses, questions, and objectives.
Identify outliers (subjects or data points with extreme values or values unlike the rest of the sample).
Examine relationships among variables.
10Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Conducting Confirmatory Analyses
Perform analyses designed to test hypothesis, research questions, or objectives.
Generalize findings from sample to appropriate populations (inference).
11Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Performing Post Hoc Analyses
Necessary when ANOVA is used in studies with three or more groups
Necessary with chi-square analyses Purpose: to determine which groups are
significantly different
12Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Probability Theory
Deductive Used to explain:
Extent of a relationship Probability of an event occurring Probability that an event can be accurately
predicted Expressed as lowercase p with values
expressed as percents
13Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Probability
If probability is 0.23, then p = 0.23. There is a 23% probability that a particular
event will occur. Probability is usually expected to be p < 0.05.
14Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Decision Theory
Inductive reasoning Assumes that all the groups in a study used
to test a hypothesis are components of the same population relative to the variables under study.
It is up to the researcher to provide evidence that there really is a difference.
To test the assumption of no difference, a cutoff point is selected before analysis.
15Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Alpha ()
Risk of making a type I error The threshold at which statistical significance
is reached
16Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Cutoff Point
Referred to as level of significance or alpha (α)
Point at which the results of statistical analysis are judged to indicate a statistically significant difference between groups
For most nursing studies, level of significance is 0.05.
Sometimes written as α = 0.05
17Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Cutoff Point (cont’d)
The cutoff point is absolute. If value obtained is only a fraction above the
cutoff point, groups are from the same population.
No meaning can be attributed to differences between the groups.
Results that reveal a significant difference of 0.001 are not considered more significant than the cutoff point.
18Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Levels of Acceptable Significance
0.05 0.01 0.005 0.001
19Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Inference
A conclusion or judgment based on evidence Judgments are made based on statistical
results Statistical inferences must be made
cautiously and with great care Decision theory rules were designed to
increase the probability that inferences are accurate
20Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Generalization
A generalization is the application of information that has been acquired from a specific instance to a general situation.
Generalizing requires making an inference. Both inference and generalization require the
use of inductive reasoning.
21Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Generalization (cont’d)
An inference is made from a specific case and extended to a general truth, from a part to a whole, from the known to the unknown.
In research, an inference is made from the study findings to a more general population.
22Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Normal Curve
A theoretical frequency distribution of all possible values in a population
No real distribution exactly fits the normal curve.
However, in most sets of data, the distribution is similar to the normal curve.
Levels of significance and probability are based on the logic of the normal curve.
23Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Normal Curve (cont’d)
24Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Tailedness
An extreme score can occur in either tail of the normal curve.
An extreme score is higher or lower than 95% of the population.
Mean scores of a population also can be extreme and occur in the tail of the normal curve.
25Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Tailedness (cont’d)
If the mean score is an extreme value, the population is not likely to be the same as that represented by the normal curve; it is significantly different.
However, extreme values that are members of the population do occur. Thus there is always a risk of making an error in deciding that the groups are different.
26Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Two-Tailed Test
Assumes that an extreme score can occur in either tail of the normal curve
Nondirectional hypothesis: tests for significance in either tail
Hypothesis: the extreme score is higher or lower than 95% of the population; thus sample with extreme score is not a member of the same population
A two-tailed test of significance is used.
27Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Two-Tailed Test (cont’d)
28Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
One-Tailed Test
Extreme values occur on a single tail of the curve.
The hypothesis is directional: one-tailed test of significance used
The 5% of statistical values considered significant will be in one tail rather than two.
Extreme values in the other tail are not considered significantly different.
One-tailed tests are more powerful than two-tailed tests.
29Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
One-Tailed Test (cont’d)
30Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Type I and Type II Errors
Type I error occurs when the researcher rejects the null hypothesis when it is true. The results indicate that there is a significant
difference, when in reality there is not. Type II error occurs when the researcher
regards the null hypothesis as true but it is false. The results indicate there is no significant
difference, when in reality there is a difference.
31Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Data analysis In reality, the In reality, the indicates: null hypothesis null hypothesis
is true: is false:
Results significant—null Type I error Correct
decision
Results notsignificant—null Correct decision Type II errornot rejected
Occurrence of Type I and Type II Errors
32Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Risk of Type I Error
33Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Power and Risk for Type II Error
Power analysis = 0.80 minimum Influenced by the sample size and the effect
size
34Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
You have six scores and the mean = 6. What is the value of score #6? Can the value of score #6 vary? Can the other five scores vary? The number of scores that can vary is your degree of freedom.
1 = 5 4 = 42 = 7 5 = 43 = 8 6 = ?
Degrees of Freedom
35Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Using Statistics to Describe
Descriptive statistics are also referred to as summary statistics.
In any study in which the data are numerical, data analysis begins with descriptive statistics.
In simple descriptive studies, analysis may be limited to descriptive statistics.
36Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Types of Descriptive Statistics
Frequency distributions Ungrouped frequency distributions Grouped frequency distributions Percentage distributions
Measures of central tendency Measures of dispersion
37Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Example of an Ungrouped Frequency Distribution
Data are presented in raw, counted form.1: /2: /////3: ///4: /5: //
38Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Example of a Grouped Frequency Distribution
Data are pregrouped into categories. Ages 20 to 39: 14Ages 40 to 59: 43Ages 60 to 79: 26Ages 80 to 100: 4
39Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Example of Percentage Distribution
Salaries: 41.7% Maintenance: 8.3% Equipment: 16.7% Fixed costs: 8.3% Supplies: 25%
40Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Commonly Used Graphic Displays of Frequency Distribution
41Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Measures of Central Tendency
What is a typical score?
42Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Mode
Is the numerical value or score that occurs with greatest frequency
Is expressed graphically Is not always the center of distribution
43Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Bimodal Distribution
44Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Median
Is the value in exact center of ungrouped frequency distribution
Is obtained by rank ordering the values When number of values is uneven, may not
be an actual value in data set
45Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Mean
Is the sum of values divided by the number of values being summed
Like the median, the mean may not be a data set value.
46Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Measures of Dispersion
Range Variance Standard deviation Standardized scores Scatterplots
47Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Range
Is obtained by subtracting lowest score from highest score
Uses only the two extreme scores Very crude measure and sensitive to outliers
48Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Difference Scores
The sum of all difference scores in a data set is zero, making it a useless measure.
Difference scores are the basis for many statistical analysis procedures.
49Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Difference Scores (cont’d)
Are obtained by subtracting the mean from each score
Sometimes referred to as a deviation score because it indicates the extent to which a score deviates from the mean
50Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Standard Deviation
Is the square root of the variance Just as the mean is the “average” value, the
standard deviation is the “average” difference score.
51Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Standardized Scores
Raw scores that cannot be compared and are transformed into standardized scores
Common standardized score is a Z-score. Provides a way to compare scores in a similar
process
52Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Scatterplots
Have two scales: horizontal axis (X) and vertical axis (Y)
Illustrates a relationship between two variables
53Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Structure of a Plot
54Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Example of a Scatterplot
55Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Chi-Square Test of Independence
Used with nominal or ordinal data Tests for differences between expected
frequencies if groups are alike and frequencies actually observed in the data
56Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Regular No RegularExercise Exercise Total
Male 35 15 50 Female 10 40 50 Total 45 55 100
Example of Chi-Square Table
57Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Chi-Square Results
Indicates that there is a significant difference between some of the cells in the table
The difference may be between only two of the cells, or there may be differences among all of the cells.
Chi-square results will not tell you which cells are different.
58Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Example of Chi-Square Results
2 = 4.98, df = 2, p = 0.05
59Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Pearson Product-Moment Correlation
Tests for the presence of a relationship between two variables Called bivariate correlation
Types of correlation are available for all levels of data. Best results are obtained using interval data.
60Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Correlation
Performed on data collected from a single sample
Measures of the two variables to be examined must be available for each subject in the data set.
61Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Correlation (cont’d)
Results Nature of the relationship (positive or negative) Magnitude of the relationship (–1 to +1) Testing the significance of a correlation coefficient
Does not identify direction of a relationship (one variable does not cause the other)
Are symmetrical
62Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Correlation Results
r = 0.56 (p = 0.03) r = –0.13 (p = 0.2) r = 0.65 (p < 0.002) Which ones are significant?
63Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Explained Variance
Definition: The R2 is the variation between two variables expressed as a percentage.
64Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Factor Analysis
Examines relationships among large numbers of variables
Disentangles those relationships to identify clusters of variables most closely linked
Sorts variables according to how closely related they are to the other variables
Closely related variables grouped into a factor
65Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Factor Analysis (cont’d)
Several factors may be identified within a data set.
The researcher must explain why the analysis grouped the variables in a specific way.
Statistical results indicate the amount of variance in the data set that can be explained by each factor and the amount of variance in each factor that can be explained by a particular variable.
66Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Usefulness of Factor Analysis
Aids in development of theoretical constructs Aids in development of measurement scales
67Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Regression Analysis
Used when one wishes to predict the value of one variable based on the value of one or more other variables
For example, one might wish to predict the possibility of passing the credentialing exam based on grade point average (GPA) from a graduate program.
68Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Regression Analysis (cont’d)
Regression analysis could also be used to predict the length of stay in a neonatal unit based on the combined effect of multiple variables such as gestational age, birth weight, number of complications, and sucking strength.
69Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Regression Analysis (cont’d)
The outcome of analysis is the regression coefficient R.
When R is squared, it indicates the amount of variance in the data that is explained by the equation.
The R2 is also called the coefficient of multiple determination.
70Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Regression Results
R2 = 0.63 This result indicates that 63% of the variance
in length of stay can be predicted by the combined effect of age, weight, complications, and sucking strength.
71Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Overlay of Scatterplot and Best-Fit Line
72Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
t-Test
Requires interval level measures Tests for significant differences between two
samples Most commonly used test of differences
73Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Example of t-Test Results
t = 4.169 (p < 0.05)
74Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Analysis of Variance (ANOVA)
Tests for differences between means More flexible than other analyses in that it
can examine data from two or more groups
75Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
ANOVA (cont’d)
Multiple versions of ANOVA are available that can be used in studies examining multiple outcome variables, or repeated measures of outcome variables across several time periods.
Can look at between-group variance, within-group variance, and total variance
76Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Results of ANOVA
F = 9.75 (2, 95) (p = 0.002) If there are more than two groups under
study, it is not possible to determine where the significant differences are.
Post hoc tests are used to determine the location of differences.
77Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Analysis of Covariance (ANCOVA)
Allows the researcher to examine the effect of a treatment apart from the effect of one or more potentially confounding variables
Potentially confounding variables that are commonly of concern include pretest scores, age, education, social class, and anxiety level.
78Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
ANCOVA (cont’d)
The effects on study variables are statistically removed by performing regression analysis before performing ANOVA.
Allows the effect of the treatment to be examined more precisely
79Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Information Needed for Algorithm
1. Determine whether the research question focuses on differences (I) or associations (relationships) (II).
2. Determine level of measurement (A, B, or C).3. Select the design listed that most closely fits
the study you are critiquing (1, 2, or 3).4. Determine whether the study samples are
independent (a), dependent (b), or mixed (c).
80Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Algorithm for Choosing a Statistical Test
81Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Judging Statistical Suitability
Factors that must be considered include: Study purpose Hypotheses, questions, or objectives Design Level of measurement
82Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Judging Statistical Suitability (cont’d)
Requires you to be familiar with the statistical procedures used in the study
Requires you to compare the statistical procedures used with other statistics that could have been used to greater advantage
Are there dependent or independent groups?
83Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Judging Statistical Suitability (cont’d)
You must judge whether the procedure was performed appropriately and the results were interpreted correctly.
Judgments required Whether the data for analysis were treated as
nominal, ordinal, or interval The number of groups in the study Whether the groups were dependent or
independent
84Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Types of Results
Significant and predicted results Nonsignificant results Mixed results Unexpected results
85Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Significant and Predicted Results
Are in keeping with those predicted by researcher and support logical links developed by researcher among the framework, questions, variables, and measurement tools
86Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Nonsignificant Results
Also called negative or inconclusive results Analysis showed no significant differences or
relationships. Could be a true reflection of reality. If so, the
researcher or theory used by researcher to develop hypothesis is in error. In this case, negative findings are an important addition to the body of knowledge.
87Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Nonsignificant Results (cont’d)
Results could stem from a type II error Causes of type II error include:
Inappropriate methods Biased or small sample Internal validity problems
88Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Nonsignificant Results (cont’d)
Inadequate measurement Weak statistical measures Faulty analysis
89Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Significant and Unpredicted Results
Are opposite of those predicted Indicate flaws in the logic of both the
researcher and the theory being tested If valid, are an important addition to the body
of knowledge
90Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Mixed Results
Most common outcome of studies One variable may uphold predicted
characteristics, whereas another does not. Or two dependent measures of the same
variable may show opposite results May be caused by methodology problems May indicate need to modify existing theory
91Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Unexpected Results
Relationships between variables that were not hypothesized and not predicted from the framework being used
Can be useful in theory development, modification of existing theory, development of later studies
92Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Unexpected Results (cont’d)
Serendipitous results are important as evidence in developing the implications of the study.
They must be evaluated carefully because the study was not designed to examine these results.
93Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Findings
Results of the study that have been translated and interpreted
A consequence of evaluating evidence
94Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Conclusions
A synthesis of the findings using: Logical reasoning Creative formation of meaningful whole from
pieces of information obtained through data analysis and findings from previous studies
Receptivity to subtle clues in data Alternative explanations of data
95Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Conclusions (cont’d)
Risk in developing conclusions is going beyond the data Forming conclusions not warranted by data Occurs more frequently in published studies than
one would like to believe
96Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Implications
The meanings of conclusions for the body of nursing knowledge, theory, and practice
Based on, but more specific than, conclusions
Provide specific suggestions for implementing the findings
97Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Significance of Findings
Associated with importance to the nursing body of knowledge
May be associated with: Amount of variance explained Control in the study design to eliminate
unexplained variance Detection of statistically significant differences
98Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Clinical Significance
Findings can have statistical significance but not clinical significance.
Related to practical importance of the findings No common agreement in nursing about how
to judge clinical significance Effect size? Difference sufficiently important to warrant
changing the patient’s care?
99Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Clinical Significance (cont’d)
Who should judge clinical significance? Patients and their families? Clinician/researcher? Society at large?
Clinical significance is ultimately a value judgment.
100Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Generalizing the Findings
Extends the implications of the findings: From the sample studied to a larger population From the situation studied to a more general
situation How far can generalizations be made?
101Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Empirical Generalizations
Are based on accumulated evidence from many studies
Are important for verification of theoretical statements or for development of new theory
Are the basis of a science Contribute to scientific conceptualization
102Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Suggesting Further Studies
Researcher gains knowledge and experience from conducting the study that can be used to design a better study next time.
Researcher often makes suggestions for future studies that logically emerge from the present study.
103Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Suggesting Further Studies (cont’d)
Replications Different design Larger sample Hypotheses emerging from findings Strategies to further test framework in use
104Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Critiquing Statistics in a Study
What statistics were used to describe the characteristics of the sample?
Are the data analysis procedures clearly described?
Did statistics address the purpose of the study?
105Copyright © 2011 by Saunders, an imprint of Elsevier Inc.
Critiquing Statistics in a Study (cont’d)
Did the statistics address the objectives, questions, or hypotheses of the study?
Were the statistics appropriate for the level of measurement of each variable?