Statistics in Experimental Stroke Research: From Sample Size Calculation to Data Description and Significance Testing
Experimental stroke researchers take samples from populations (e.g., certain mouse strains), and make inferences about unknown parameters (e.g., infarct sizes, outcomes). They use statistics to describe their data, and they seek formal ways to decide whet
- PDF / 227,526 Bytes
- 13 Pages / 504 x 720 pts Page_size
- 94 Downloads / 161 Views
Experimental stroke researchers produce data: infarct volumes, cell counts, functional outcomes and such. Collecting data means sampling from a population. Since the population (e.g., all SV 129 mice) is too numerous to be sampled, we collect a sample to represent the population, and we use our sample (e.g., 20 SV 129 mice) to make inferences about unknown parameters of the population: infarct sizes, the effect of treatment with “Compound Ulrich Dirnagl (ed.), Rodent Models of Stroke, Neuromethods, vol. 47, DOI 10.1007/978-1-60761-750-1_18, © Springer Science+Business Media, LLC 2010
249
250
Dirnagl
X,” etc. Descriptive statistics help us to do just this. But very often stroke researchers also want an “objective method” to decide whether an observation from a sample justifies accepting a hypothesis about the population. In other words, scientists want to test hypotheses and make decisions. In statistical terms, asking whether Compound X reduces infarct sizes is equivalent to asking whether the two samples (control vs. Compound X) come from the same population. Unfortunately, there is a certain likelihood that our decision might be wrong, as our test statistic could result in a false positive or a false negative. To get a handle on these errors, we calculate p values for levels of significance and power. P values, in particular p = 0.05, have become the graven image of modern biomedicine. Most stroke researchers believe that the p-value of a significance test is the probability that the research results are due to chance and are completely unaware of the concept of statistical power. The overall aim of this chapter is to guide the preclinical stroke researcher in the proper experimental planning and statistical evaluation of data. Before presenting an example of a mock study, from hypothesis to significance testing, a few concepts need to be clarified. The following section should serve as a primer in “Statistics for the experimental stroke researcher.”
2. Concepts 2.1. Descriptive Statistics
Types of data in stroke research: Data might be numerical, such as infarct volume data, or categorical, such as functional score data. 1. Numerical variables tell us: “how much, how many?”, and can be either continuous (e.g., infarct volumes) or discrete (e.g., hemorrhagic transformation count). Numerical variables can be treated with mathematical operations. Numerical summaries of quantitative data can be used to describe the location (e.g., mean) and the spread (e.g., variance) of the data, and statistical analysis may use parametric tests (e.g., t-test, ANOVA), provided the data is normally distributed. 2. Categorical variables tell us: “what type?”, and can be nominal (unordered; e.g., male, female) or ordinal (ordered; e.g., Bederson outcome score: 0,1,2,3). A categorical variable places an individual into one of several categories. Mathematical operations make no sense with them. They can be numerically summarized with a count or with relative frequencies. Categorical variables can be visualized with dot plots or bar
Data Loading...