Glossary terms for 'M'
|Marginals|| The row and column totals of a contingency table. For example, looking at the marginals in the 2 x 2 table showed that there were similar numbers of men and women in the study.|
|Matching||In a case-control study, the process of selecting controls to be similar in certain attributes to cases, to reduce confounding by those attributes. For example, in a case-control study of the risk factors for brucellosis, controls were matched to cases by age (within 3 years), sex, and county of residence. See also overmatching. |
|Mean||The average value of a continuous variable in a sample or population; calculated as the sum of all the values of that variable divided by the number of subjects. For example, the mean serum cholesterol level in a sample of 287 middle-aged women was 223 mg/dL. See also median and standard deviation. |
|Measurement error||The situation in which the precision or accuracy (or both) of a measurement is less-than-perfect; thus there is at least some measurement error for most variables (with the possible exception of death). For example, to reduce measurement error, the investigator used a 2 kg stainless steel weight to calibrate the baby scale weekly.|
|Median||The value of a variable that divides a sample or population into two halves of (approximately) equal size; equivalent to the 50th percentile. Often used when a continuous variable has a few very high (or very low) values that would overly influence the mean value. For example, the median annual income in the sample of 54 physicians was $225,000. See also mean and standard deviation. |
|Medical test studies||A general term used for studies that measure how well a test (or a series of tests) identifies patients with a particular diagnosis or outcome. For example, the investigator performed a medical test study to determine the likelihood ratios for the presence and absence of typical angina (defined as exertional substernal chest pain or pressure) for the diagnosis of coronary artery disease. |
|Meta-analysis||A process for combining the results of several studies with similar predictor and outcome variables into a single summary result. For example, a meta-analysis of 12 published studies found that use of non-steroidal anti-inflammatory drugs was associated with a 28% greater risk of developing asthma.|
|Missing data||Data that were not collected during a study, whether at baseline or during follow-up. For example, the investigator was concerned that the relatively large proportion (34%) of subjects who had missing data on alcohol use may have biased her study of the risk factors for falls.|
|Multiple hypothesis testing||The situation in which an investigator studies more than one?and usually many more than one?hypothesis in a study, thereby increasing the risk of making a Type I error (see below) unless the level of statistical significance (see alpha, above) is adjusted. For example, although the investigator reported a statistically significant (P = 0.03) association between consumption of Vitamin D and cognitive decline, her results were criticized because she did not account for the effect of multiple hypothesis testing, since the study had looked at more than 30 nutritional supplements. See also Bonferroni correction.|
|Multiple-cohort study||A cohort study (see above) that enrolls two or more distinct groups of subjects (the cohorts), and then compares their outcomes. Often used in studies of occupational exposures, in which the cohorts being compared are either exposed to a potential risk factor or not. For example, the investigators performed a multiple cohort study of whether exposure to cosmic rays during airplane flights is associated with an increased risk of hematologic malignancies; the investigators studied four cohorts: pilots and flight attendants (who would be exposed to cosmic rays) and ticket agents and gate attendants (who would not). |
|Multivariate adjustment||A general term for the statistical techniques used to adjust for the effects of one or more potential confounding variables on the association between a predictor and outcome. For example, using multivariate adjustment, the study found that ingestion of supplemental Vitamin D was associated with an increased risk of cognitive decline, adjusting for age, sex, education, baseline cognitive function, and smoking.|
Glossary material from Hulley SB et al. Designing Clinical Research, 4th ed. Philadelphia, Lippincott Williams & Wilkins, 2013.