Short answers, pulled from the story.
Statistical bias is a systematic tendency in data collection and analysis methods that produces an inaccurate, skewed, or distorted depiction of reality. It matters because data is used to inform lawmaking, industry regulation, corporate marketing, and institutional policies, so unchecked bias can have significant real-world consequences.
The main types include selection bias (also called Berksonian bias or sampling bias), volunteer bias, funding bias, attrition bias, and recall bias. Each arises at a different point in the data collection process and can distort results in distinct ways.
A Type I error, or false positive, occurs when the null hypothesis is correct but is rejected. A Type II error, or false negative, occurs when the null hypothesis is incorrect but is accepted. A test that reduces one type of error often increases the other.
Biased estimators are sometimes preferred because an unbiased estimator may not exist without additional assumptions, may be difficult to compute, or may have a higher mean squared error than the biased alternative. In the case of the Poisson distribution, a biased estimator is always positive and has a smaller mean squared error than the corresponding unbiased estimator.
Spectrum bias arises when diagnostic tests are evaluated on patient samples that are not representative of the general population. A high prevalence of disease in a study population inflates positive predictive values, creating a gap between predicted and actual values in a broader clinical setting.
Researchers can reduce bias by defining clear research parameters from the start, using blind or double-blind techniques to address observer bias, avoiding p-hacking, and rerunning analyses with different independent variables to check whether findings hold. Careful language in reporting, such as avoiding the phrase "approaching significance" for results that did not achieve it, also helps prevent misleading interpretations.