Prof. Umasankar Mohanty, B.P.T, M.P.T (Manual Therapy), F.A.G.E

The present day health care delivery system believes in evidence-based practice. So a strong scientific temperament and research is necessary for any healthcare system for establishing the treatment methods. The field of physiotherapy has substantial scientific temperament. In order to establish a fact into a true science we rely on experimentation, data collection and statistical analysis. Research methodology and statistical components lends credibility, not only academically but professionally. It is important to understand research methods and statistics to evaluate other professionals research activities and reports as well as to put our own research and conclusions.  

Statistics is not a magic act. At its core, is the effort to understand variation. Statistics helps us to understand variation and its correlations. It facilitates us to derive some conclusion to our experimentation. In the article we put a birds eye view on the different statistical methods adapted to derive into some conclusions. In this overview I’ll provide some definitions of basic statistics terminology and attempt to provide a flow chart that helps to understand the type of information provided by some common statistical tests. I am by no means a statistician; my goal is for a conceptual understanding of the subject not mathematical.

Research is defined as a detailed study of a subject in order to discover (new) information or reach a (new) understanding. When engaged in research we must always end up in measuring something (muscle tone, pain assessment, functional improvements). These measurements are called as Data. There are two main ways of data presentation in research designs

RESEARCH DESIGN

The inferential statistics allow the researcher to make assumptions beyond the set of data in front of her/him.
The inferential statistics is presented with 2 designs a. Experimental designs b. Correlational designs

Attempts to define a cause and effect relationship through group comparisons. The experimental design is of following types:

A-B design: Involves two phases, a pretreatment or baseline phase followed by an intervention or treatment phase.
A-B-A design: Multiple baseline design involving three phases (baseline phase, treatment phase, second baseline phase)
A-B-A-B design: Multiple baseline, Multiple treatment includes baseline, treatment, additional baseline and treatment phases.

Retrospective - Investigation data collected in the past.
Prospective - Recording and investigation of present data.
Predictive - Useful to develop predictive models.
Attempts to determine whether a relationship exists between two or more quantifiable variable and to what degree.
Degree of relationship is expressed as correlational coefficient, ranging from –1.00 to +1.00.
If the correlation near +1.00, the variables are positively correlated.
If the correlation near 0.00, the variables are not related.
If the correlation near -1.00, the variables are inversely related.

VARIABLES

HYPOTHESIS     
               
A tentative and testable explanation of the relationship between variables; the results of an experiment determine if the hypothesis is accepted or rejected.

LEVELS OF MEASUREMENT

There are four levels of measurement. These levels are a hierarchy with nominal at the lower end and ratio at the top. As you move from higher to lower you lose information and have less ability to mathematically and statistically manipulate the data.

Nominal: Qualitative data with no natural ordering; Characteristics or attributes may be randomly coded numerically but the numbers really provide no information about the variable. For instance you might have the categories male and female and code them "1" and "2" respectively.

Ordinal: Qualitative data with an ordering. For instance if you were to measure ADL’s using a scoring system of 1=dependent, 2=max assist, 3=mod assist, 4=min assist, 5=independent.

Interval: Quantitative data in which there is a clear ordering and the distance between objects is specified but there is no meaningful zero point (absence of attribute). For instance temperature measured in Fahrenheit.

Ratio: Quantitative data which have classifies and ranks variables or scores based on equal intervals and a true zero point; the highest, most precise level of measurement; e.g. goniometry, scales for height, weight, or force allow the use of precise physical measures for research.

SAMPLING

The selection of individuals (a sample) for a study from a population; the sample represents the largest group from which they were selected.

DATA ANALYSIS AND INTERPRETATION

For Descriptive Statistics
Measures of Central Tendency

Mode: The numeric value that occurs most frequently.
Median: The point on the numeric scale above which and below which 505 of the cases fall.
Mean: Sum of scores divided by total number of scores, arithmetic average.

Measures of Variability

Range: highest score minus the lowest score.
Standard deviation: used with interval or ratio level data, summarizes the average amount of deviation of all values of a distribution from the mean.
Variance: value of the standard deviation before a square root is taken. Variance is not typically reported in research but is often used in inferential statistical tests.

For Inferential Statistics

Allows the determination of how likely the results of a study of a sample can be generalized to the whole population.

Alpha level: Preselected level of statistical significance. Most commonly .05 or .01: indicates that the expected difference is due to chance, e.g at .05, only 5 times out of every 100 or a 5%chance, often expressed as a value of p.

Parametric vs. Nonparametric tests: Parametric tests require measurement on at least an interval scale and typically have a number of other assumptions regarding the distribution of variables. Parametric tests are measured with T-Test, Analysis of variance (ANOVA) and Analysis of covariance (ANCOVA). Nonparametric tests are applied to data collected on nominal or ordinal scales and have relatively far fewer assumptions. Non parametric tests are measured with Chi square test.

Answers to some frequently asked questions

How well does one variable predict another? Regression analysis
What is the degree that two variables are related?
Ratio / Interval data? Pearson Product Moment Correlation
Ordinal data? Spearman Rank Correlation
Is there a difference between/among group means?
Nominal data? Chi square
Ordinal data? Kruskal-Wallis One-way ANOVA by ranks (if less than 2 groups / comparisons) Wilcoxon Signed-Ranks test (more than 2 groups / comparisons, paired data)

Mann-Whitney U Test or Wilcoxon Rank Sum test (more than 2 groups / comparisons, unpaired data)
Ratio / Interval Data? T-Test (if only 2 groups / comparisons, may be done paired or unpaired)
ANOVA (if more than 2 groups / comparisons)

References