Essential Statistics for Nursing Research Kristen Carlin, MPH Seattle Nursing Research Workshop January 30, 2017 Table of Contents Plots Descriptive statistics Sample size/power Correlations Hypothesis testing Time series data Interpreting statistics in practice Plots 1
Types of Plots Scatterplot Boxplot Histogram Why should I create plots? Provides a visualization of the data Helps to understand relationships between variables Show the shape of the data How are the data distributed? 2
Scatterplot Show how much one variable is affected by another The closer the data points come to making a straight line, the higher the correlation (or linear association) between the two variables 3
Boxplot Displays the distribution of data Based on the 5 number summary Minimum 1 st quartile (25 th percentile) Median (50 th percentile) 3 rd quartile (75 th percentile) Maximum Histogram Different than a bar chart Bar charts display categorical data Histograms display quantitative data Displays shape of the distribution 4
Descriptive Statistics What are descriptive statistics? Numbers used to summarize and describe data Provide information about the study sample 5
Types of Descriptive Statistics Central Tendency What is the middle of the data like? Mean Median Mode Statistical Dispersion How spread out are the data from the middle? Standard Deviation Interquartile Range Range Measures of Central Tendency Mean Average of the measurements Sum/number of measurements Median Number at which half your measurements are more and half are less than that number Mode Measurement that has the greatest frequency Measures of Dispersion Interquartile Range Difference between the third quartile and the first quartile Range Difference between the largest measurement and the smallest one Variance Measurement of how far values are spread out Average of the squared differences between each value and the mean Small = data points are close to the mean and to each other Large = data points are very spread out Standard Deviation Square root of the variance 6
Sample Size & Power Calculations Why bother calculating a sample size? A study without adequate power may not be able to detect interesting effects Many funding sources require sample size calculations to ensure that studies are worthwhile Some journals want to know that the study was adequately powered to detect reasonable effects prior to publication 7
Correlations What is correlation? Describe the linear association between two quantitative variables Correlation does not mean causation Correlation coefficient ranges from -1 to 1 R 2 is the proportion of variability in one variable explained by the other Value of Correlation Coefficient (r) Interpretation - 1 Strong negative correlation Close to 0 Weak correlation + 1 Strong positive correlation 8
What is your interpretation of the correlation coefficient (r)? Remember: Correlation does not mean causation. Hypothesis Testing 9
Parametric vs. Non-Parametric Parametric Normal distribution Non-Parametric Non-Normal distribution Which test should I choose? Think through your research question Is your data parametric or non-parametric? Are you comparing pre-and-post scores on the same subjects? Paired t-test, Wilcoxon signed rank Are you comparing two groups? T-test, Wilcoxon rank sum Do you have multiple groups? ANOVA, ANCOVA, Non-parametric ANOVA with Kruskal-Wallis test Are you comparing categorical data? Chi-square Are you testing the association between two quantitative variables? Correlation, linear regression Consult with a statistician What is a confidence interval? Estimate of Interest 10
Time Series Data What is time series data? Sequence of data points made over time Cross-sectional data cannot be used to evaluate causation Allows us to remove confounding effect of time Examples: Quality improvement projects Nurse sensitive indicator data 11
Misleading Statistics Study Sample Why might proper sampling be important? We cannot (usually) obtain data on an entire population Cost could be enormous Not all eligible people will consent Some people will still fall through the cracks Sample should be generalizable to the population you want to study 12
Graphs What are some factors that may cause graphs to be misleading? Scale Does it start with a value that makes sense given the data? (ie, zero) Are numbers equally spaced? Title Do the titles/captions make sense? Source of the data Where is the data from? Graph (Example) 7 6 5 4 3 2 1 0 6 5 4 This is the SAME data, just different scales. Issues with Design and Analysis Lacking scientific rationale when developing a research question and hypotheses 13
Issues with Design and Analysis Choosing a parametric test without assessing for normality of the data Not controlling for confounders in the analysis Age Activity level Weight gain Not adjusting for multiple comparisons Critically Interpreting Results Relying solely on the p-value Clinically relevant findings are not always statistically significant Statistically significant findings are not always clinically relevant What was the effect size? (ie, what was the difference between groups?) Look at confidence intervals 14
Okay, on your own now Describe the study sample Describe the statistical analysis methods Describe the main findings Where did the authors do a great job? Where could they have improved? One more reminder A non-statistically significant effect may be clinically important A statistically significant effect may not be clinically important 15
Questions? 16