STATISTICAL ANALYSIS 101

Dr. Marla Kniewel

Nebraska Methodist College
 OBJECTIVES

• Distinguish descriptive from inferential statistics

• Apply the decision path in determining statistical tests to use in data analysis

• Determine appropriate parametric or nonparametric statistical tests to use in

data analysis
 Research Purpose

Describe data Examine differences Examine relationships Predict relationships

2 Groups Correlation Regression Analysis

Frequencies Pre-test / Post-test Statistic
Percentages - Linear Regression
- t-test - Pearson’s r
Means (SD) - Multiple regression
- Mann-Whitney U test - Spearman Rho
- Logistic regression
- Wilcoxen - Kendall’s Tau
- Chi-Squared - Chi-Square

> 2 Groups
- ANOVA
- ANCOVA
- MANOVA
Pre-test / Post-test
- RM-ANOVA
 LEVELS OF MEASUREMENT

Nominal Ordinal Interval Ratio

• Gender • Pain scale (0-10) • Temperature • Age

• Ethnicity • Age groups (18-25, 26- • IQ • Height
• Marital status 35, etc.) • SAT score • Weight
• Zip code • Grade (A, B, C, D, & F) • Depression score • BP
• Religious affiliation • Satisfaction scale (poor, • Time of day • HR
• Medical diagnosis acceptable, good) • Dates (years) • Years of experience
• Names of medications • Performance scale • Time to complete a task
(Below average, average,
above average)
 CATEGORIES OF STATISTICS

• Descriptive Statistics • Inferential Statistics

• Describe situations and events • Allows conclusions about variables
• Summary (numbers, percentages) • Statistical tests are performed
• Central Tendency • Comparisons
• Charts / Graphs • Associations
• Predictions
 DESCRIPTIVE STATISTICS

• Describe, Summarize & Organize data

• Frequency distributions
• Graphs/Tables
• Measures of Central Tendency/Dispersion
• Mean (M)
• Standard deviation (SD)
x
x

DESCRIPTIVE STATISTICS

• Distribution of data

• Normal distribution

• Skewness

• Negative skew

• Positive skew
 INFERENTIAL STATISTICS

• Probability • Examine relationships among variables

• Likelihood an outcome will occur • Correlation statistics
• Helps identify risk
• Confidence Interval (CI) • Predict relationships among variables
• Regression analysis
• Alpha level (α-level) or significance level
• Defines statistical significance • Examine / Compare differences between variables
• Most common in healthcare: .05 and .01 • t-test
• p-value • ANOVA
 PARAMETRIC VS NONPARAMETRIC

PARAMETRIC STATISTICAL NONPARAMETRIC STATISTICAL

TESTS TESTS
• Assumptions • No assumptions of distribution
• Data must be normally distributed • Small sample size
• Interval or ratio data • Level of measurement
• Independence of data • Nominal or ordinal
• Need sample size >30
• More powerful
 WHAT STATISTIC AL TEST TO USE?
PARAMETRIC OR NONPARAMETRIC

Example:
 Sample of critically ill patients
 Length of stay
 20 females
 Mean = 60
 Median = 31.5
 19 males
 Mean = 30.9
 Median = 30
 EXAMINE RELATIONSHIPS

• Correlation Statistics
• Exploratory studies
• Examines relationship between variables
• Direction of relationship
• Doesn’t specify IV & DV

EXAMPLES
• Association between overtime hours worked and medication errors in RNs
• Relationship between social support and stress in elderly rural women
• Relationship between time on ventilator and LOS in ICU patients
 CORRELATION COEFFICIENTS

• Direction of relationship

• Strength of relationship (-1 to +1)

• ≥ .10 → weak
• ≥ .30 → moderate
• ≥ .50 → strong

• Statistical significance
 PEARSON’S CORRELATION
COEFFICIENT (r)

• Parametric test
• Assumptions
• Normal distribution / Interval or ratio
• Related pairs / Absence of outliers
• Linearity / Homoscedasticity
• Interval or ratio data level
• Reported as:
• r = .78, p < 0.001
 NONPARAMETRIC TESTS

• Spearman Rho • Kendall’s Tau Chi Square test

• Skewed distribution • Skewed distribution  One variable- nominal
• One variable- ordinal • One variable- ordinal  No direction or
level level association reported
• Reported as: • Reported as:  Reported as:
• rs = .82, p = .042 • r t = .82, p = .042  χ2 (1)= 5.00, p = .025
Good, H., Riley-Doucet, C., & Dunn, K. (2015). The prevalence of uncontrolled pain in long-term care: A pilot study examining
outcomes of pain management processes. Journal of Gerontological Nursing, 41(2), 33-41.
 PREDICT RELATIONSHIPS

• Regression Analysis
• Exploratory & Prediction studies
• Quantifies a relationships among variables to predict future events
• Estimates values for DV by known values of IV
• Dependent variable (DV) – outcome variable
• Independent variable (IV) – influencing variable
• Makes inferences or predictions
• Statistically significant correlations (≥ .50)
• Measure strength of association
 3 TYPES OF REGRESSION ANALYSIS

• Linear regression
• Relationship between a single independent variable and a single interval- or ratio-level variable
• Predicts the future value of dependent variable based on level of independent variable
• Results report: R and R2
• Multiple regression
• Make prediction about how 2 or more independent variables affects the dependent variable
• Reported as R2
• Logistic regression
• Used when dependent variable is categorical (nominal or ordinal with 2 categories)
• Generates an Odds Ratio (OR)
 EXAMPLES
Investigate the relationship between gestational age at birth (weeks) & birth weight (lbs.)
• Simple linear regression
• Significant relationship between gestation and birth weight (r = .706, p < 0.001).
• Slope coefficient for gestation was 0.355
• Weight of baby increases by 0.355 lbs. for each extra week of gestation.

Investigate the effect of age (years) and height (inches) on weight

• Multiple regression
• Significant relationship between age and weight (r = .476, p = .001); height and weight (r = .672, p = .001)
• Control for height, every year adds 1.71 lbs.
• Control for age – every inch adds 10.37 lbs.
 EXAMPLE:
LOGISTIC REGRESSION
• Purpose: Identify factors that can be related OR CI p
to the occurrence of gestational arterial Race White 1
hypertension.
Non-white 8.18 1.39-48.10 0.020

• Dependent variable Depression No 1

• Momentary HTN – yes
• Momentary HTN – no Yes 8.69 1.19-63.42 0.033

Obesity No 1
• Independent variables
• Anxiety Yes 6.45 1.40-29.61 0.016
• Depression Anxiety No 1
• Obesity
• Demographic variables Yes 7.77 1.19-50.45 0.032
• Age Franco, R., Ferrreira, C., Vieira, C., & Silva, R. (2015). Ethnicity, obesity, and
emotional factors associated with gestational hypertension. Journal of Community
• Race Health, 40(5), 899-904. DOI: 10.1007/s10900-015-0010-8
• Education
 EXAMINE DIFFERENCES

• Experimental Designs • Statistical test to use depends on

• Effect of IV on DV • # of groups
• 1 or more variables • Level of measurement
• 1 or more groups • Type of sample
• Comparing results (means) • Independent samples
• Between subjects • Dependent samples
• Within subjects
 PARAMETRIC TEST:
Differences between 2 means

• Students t-test One-sample t–test

• One-sample t–test • Interval or Ratio data
• Independent Samples t-test • Compare mean to known value
• Dependent Samples t-test • Results reported as
t (99) =2.224, p =0.028
• One-tailed or two-tailed test
 PARAMETRIC TEST:
Difference between 2 means
INDEPENDENT SAMPLES DEPENDENT SAMPLES
t-test t–te st
• Interval or Ratio data • Paired t–test
• Independent samples • Interval or Ratio data
• Results reported as • Dependent samples
t (18) =2.86, p =0.011 • Results reported as
t (15) = 4.00, p = 0.001
 EXAMPLES

70 patients with leukemia Experimental Group Control Group

(n=35) (n=35)
• Experimental group (n=35)
m SD m SD t
• 2 follow-up phone calls (IV)
• Control group (n=35) Self-care 2.67 .036 1.78 0.38 10.347*

• Routine care *p<.001

• Self-care (DV)
Pre- class Post- class

70 patients with hypertension m SD m SD t

• Stress reduction classes
SBP 178.92 24.6 131.47 6.38 4.467*
• SBP (pre & post)
*p<.001
 NONPARAMETRIC TESTS:
Differences between 2 medians

MANN-WHITNEY U-TEST WILCOXON RANKED-SIGN TEST

• Looks at differences in distribution of a variable • Looks at differences in distribution of a variable
• Assumptions • Assumptions
• Random samples • Random samples
• Independent samples • Dependent samples
• Level of measurement: Ordinal +
• Level of measurement: Ordinal +
• Results of test are reported as
• Results of test are reported as
• U = 67.5, p = .034
• (Mdn =21.00), Z = -1.807, p = 0.071, r = -.84
• Wilcoxon Rank-Sum test
• Wilcoxon Matched-Pairs test
• Ws = 109.50, p = .008
 NONPARAMETRIC TESTS:
Differences between 2 frequencies

CHI-SQUARE ( X 2) STATISTIC
• Looks at differences in distribution of
frequencies
Do not use Use
• Level of Measurement: nominal or ordinal
• Independent groups < 30
Video
105
Link
Antihistamines Antihistamines
32
Total
137
> 30 (skip through
72 the math part)
9 81
• Observed frequencies vs. Expected
frequencies Total 177 41 218

• Results reported as
• X 2 (2, N = 218) = 14.14, p < 0.01
 ANALYSIS OF VARIANCE (ANOVA)

• Parametric Test Types of ANOVAs

• Differences in means between >2 Groups
• One-way ANOVA
• Post hoc tests
• Bonferroni • ANCOVA
• Tukey’s
• Two-way ANOVA
• Scheffé’s
• N-way (Factorial) ANOVA
• Reported as an F ratio
• F(59, 56) = 7.77, p = .042 • RM-ANOVA

• MANOVA
 ONE-WAY ANOVA

 3 or more Independent Groups • Assumptions

 Comparing 3 or more means • Normal distribution
• DV at least Interval level
 1 independent variable (1 factor) • Variances in groups are same

 1 dependent variable • Independent samples

 Results reported as
 F(2, 27) = 4.98, p = 0.01
 EXAMPLE using ONE-WAY ANOVA

Group A Group B Group C

Smokers Counseling Nicotine Patch Chantix
• IV- methods to quit smoking 16.6 19.2 34.0
n=8 n = 10 n=9
• DV- # cigarettes/day
F (2, 27) = 4.98, p = 0.01

• Ho: There is no significant difference in number of cigarettes per day between smokers who had
counseling, used a nicotine patch, or used Chantix.
• Ha: There is a significant difference in number of cigarettes per day between smokers who had
counseling, used a nicotine patch, or used Chantix.
 ANCOVA (Analysis of Covariance)

 3 or more Groups • Assumptions

 Comparing 3 or more means • Normal distribution
• DV at least Interval level
• Variances in groups are same
 1 Independent Variable (factors)
• Independent samples
 1 Dependent Variable • Independence between covariate & IV
• Relationship between covariate & DV stays the
same
 Adjusts scores on dependent
variable
 Removes effect of confounding variables
(covariates)
 EXAMPLE using ANCOVA

EXERCISE STUDY

• IV- Exercise • DV- Health Problem Index

• No exercise • Confounding variable: weight

• Exercise 1x / week

• Exercise 3x’s / week

• Exercise 5x’s / week

 EXAMPLE USING TWO-WAY ANOVA

• Smokers study H01: There is no significant difference in the mean #

cigarettes/day among participants getting counselling or the
• IV #1- Methods to quit smoking (Factor A) nicotine patch.
• IV #2- Gender (Factor B) H02: There is no significant difference in the mean #
cigarettes/day among male or female participants.
• DV- # cigarettes/day
H03: There is no significant interaction effect between gender
and method used..
 RM-ANOVA (Repeated Measures ANOVA)

 Dependent Sample
 Comparing 3 or more means
Assumptions
 DV measured at different times
 >2 IV to same group
• Normal distribution
• DV at least Interval level
 1 or more Independent Variables • Variances in groups are same
 1 or more Dependent Variable • Sphericity

 Results reported as
 F(1, 108) = 3.93, p < 0.001
 RM-ANOVA EXAMPLES

Example 1 Example 2
n = 100
Video Game Music Television
(Weeks 1-4) (Weeks 5-8) (Weeks 9-12)
9.4(1.96) 16.9(3.05) 13.8 (2.63)
DV: Quality of sleep
Example 3
n = 100
Video Game Music Television
(Weeks 1-4) (Weeks 5-8) (Weeks 9-12)
Male 12.4(2.16) 16.9(3.05) 13.8 (2.63)
Female 14.9 (3.21) 18.3(2.78) 15.5 (1.45)
DV: Quality of sleep
 NONPARAMETRIC TESTS:
Kruskal-Wallis

• Compares medians
• Any assumption of ANOVA testing is violated
• Small sample size
• Assumptions
• Random samples
• Independent samples
• Level of measurement: Ordinal, Interval, or Ratio
• Only tells that measurements differ
• Mann-Whitney test provides which differ significantly
 NONPARAMETRIC TESTS:
Friedman’s ANOVA

• Nonparametric equivalent for RM-ANOVA

• Uses ranked data
• Any assumption of ANOVA testing is violated
• Small sample size
• Assumptions
• Independent measurements
• Level of measurement: Ordinal, Interval, or Ratio
• Only tells that measurements differ
• Wilcoxon Signed-rank test provides which differ significantly
 STATISTICAL SIGNIFICANCE vs.
CLINICAL SIGNIFICANCE

Statistical Significance Clinical Significance

• Results unlikely to caused by chance • Magnitude of risk reduction

• Dependent upon Power and # of subjects in a • Precision of the treatment effect

study • Effect Size (ES)
• Cohen’s d calculation
• The larger the sample the greater the power • Confidence Interval (CI)
& probability of detecting significant results
between variables
 EFFECT SIZE CALCULATION-
COHEN’S d

mean1-mean2 • Small ES ≥ .20

SD1 + SD2/ 2
• Medium ES ≥ .50

• Large ES ≥ .80
EXAMPLE OF CLINICAL SIGNIFICANCE