Meta-analysis

Scientists often ask the same big questions.

They do many small studies to find answers. One study might not show the whole truth. A meta-analysis looks at many studies at once. It puts all those facts together. This helps scientists see a clearer picture. It can fix mistakes from just one study. This method is used in medicine and science. It helps leaders make good rules for health. It is a powerful way to learn more.

Using many studies makes our knowledge much stronger.

Scientists often ask the same big questions. They do many small studies to find answers. One study might not show the whole truth. A meta-analysis looks at many studies at once. It puts all those facts together. This helps scientists see a clearer picture.

This method can fix mistakes from just one study. It uses math to find a combined effect size. This size shows the true result across all studies. Meta-analysis helps make health rules and new laws.

It can even compare two treatments indirectly. This happens when we compare both to a third group. A man named Gene Glass named this method in 1976. He called it the analysis of analyses. In 1904, Karl Pearson used an early version. He looked at data from studies on typhoid. This work helped people learn about medicine. Today, many fields use this tool. It helps in medicine, psychology, and ecology. The number of these studies has grown fast. There were 334 in 1991. By 2014, there were 9,135.

This growth shows how much we value this way of learning.

Scientists often ask the same big questions. They run many small studies to find answers. One study might not show the whole truth. A meta-analysis looks at many studies at once. It puts all those facts together. This helps scientists see a clearer picture.

This method can fix mistakes from just one study. It uses math to find a combined effect size. This size shows the true result across all studies. Meta-analysis helps make health rules and new laws. It also helps guide future research projects.

To start, a researcher must collect data. They search through large databases like PubMed or PsychInfo. They use specific keywords to find the right papers. Sometimes they use a trick called snowballing. This means looking at the reference lists of good papers to find even more.

They must be very careful to pick only the right studies. They often use a flow diagram to show how they picked them. This keeps the work honest and easy to repeat.

People have used these ideas for a long time. A statistician named Karl Pearson used an early version in 1904. He looked at data from studies about typhoid inoculation. Later, a man named Gene Glass coined the term "meta-analysis" in 1976. He said it is the "analysis of analyses."

The first modern model was published in 1978. It looked at how well psychotherapy worked for people.

Not everyone liked the idea at first. In 1978, Hans Eysenck called meta-analysis "mega-silliness." He even called it "statistical alchemy." But the method became very popular anyway. In 1991, there were only 334 published meta-analyses. By 2014, that number grew to 9,135.

Today, people use it in medicine, psychology, and even ecology.

There are different ways to do the math. One way uses aggregate data, which is a summary of a study. Another way uses individual participant data, which is the raw information.

Researchers can use a "fixed effect model" if the studies are very similar. If the studies are different, they use a "random effects model." This model accounts for the differences between the groups. This helps ensure the final answer is as accurate as possible.

Meta-analysis is a specialized statistical method used to synthesize quantitative data. It brings together findings from multiple independent studies that all address a single research question. Instead of looking at just one experiment, researchers extract effect sizes and variance measures from many different sources. By combining these effect sizes, scientists can calculate a single, combined effect size. This process increases statistical power and helps resolve uncertainties or disagreements found in individual studies.

Consequently, meta-analyses are essential for supporting research grants and shaping medical treatment guidelines. They also influence health policies and help guide the direction of future scientific studies.

The process begins with an intensive literature search to collect data. Researchers must identify appropriate keywords and search limits to navigate databases like PubMed, Embase, or PsychInfo. Many scientists use duplicate search terms across multiple databases to ensure they do not miss important information. They might also use a technique called "snowballing," which involves searching the reference lists of eligible studies for more relevant papers. Once a large volume of studies is found, researchers must filter them carefully. They use pre-specified criteria to discard studies based on their titles or abstracts.

To maintain transparency, researchers use a PRISMA flow diagram to document how many studies were found and why certain ones were removed.

Data collection requires a standardized approach, often using a specific data collection form. For studies involving correlational data, researchers typically collect the Pearson's r statistic. They must be careful with partial correlations, which can sometimes inflate relationships compared to zero-order correlations. If necessary, researchers can use plot digitizers to scrape data points directly from scatterplots. Beyond the numbers, it is vital to collect study characteristics that might moderate effects, such as the mean age of participants. Researchers also assess the quality of each study using various tools to check for potential bias.

They may even include "gray literature," which is research like dissertations or conference abstracts that has not been formally published.

There are two primary types of evidence used in these analyses: individual participant data (IPD) and aggregate data (AD). Aggregate data consists of summary estimates, such as odds ratios or relative risks, which are commonly found in published literature. This data can be direct or indirect. Indirect aggregate data allows researchers to estimate the effect between two treatments that were never compared directly. For example, if Treatment A and Treatment B were both compared to a placebo in separate studies, researchers can estimate the difference between A and B.

In contrast, IPD represents the raw, original data collected by study centers. This distinction has led to the development of one-stage methods, which model all raw data simultaneously, and two-stage methods, which calculate summary statistics first.

Statisticians use different mathematical models depending on the nature of the studies. The fixed effect model provides a weighted average of study estimates. In this model, the weight is usually the inverse of the estimate's variance, meaning larger studies contribute more to the final result. This model assumes that all included studies investigate the same population and use identical definitions for variables. However, this assumption is often unrealistic due to heterogeneity, which refers to the variability between studies. If a meta-analysis is dominated by one massive study, the findings from smaller studies may be practically ignored under a fixed effect approach.

To account for differences in study methods or sample characteristics, researchers often use a random effects model. This model treats heterogeneity as a random component. It uses a two-step process: first, it applies inverse variance weighting, and second, it "un-weights" that value using a random effects variance component (REVC). This component is derived from the extent of variability among the effect sizes. If the variability is very high, the model may eventually treat all studies with equal weight.

While useful, the random effects model can sometimes lead to overconfident conclusions or underestimate statistical error.

The history of this field shows a journey from early data collation to a formal science. While the term "meta-analysis" was coined by statistician Gene Glass in 1976, the practice is older. In 1904, Karl Pearson published a paper in the British Medical Journal that aggregated data from typhoid inoculation studies. This is considered one of the first uses of a meta-analytic approach in clinical studies. Modern meta-analysis truly took shape with the 1978 work of Mary Lee Smith and Gene Glass regarding psychotherapy. Despite early criticisms, such as Hans Eysenck calling the method "statistical alchemy," the field has exploded. The number of published meta-analyses grew from 334 in 1991 to 9,135 by 2014.