CentralStatz Statistical & Data Sciences LLC

In this video we go through the R code that facilitates our blog post titled, "A Simple Example Why Statistical Significance Is Insufficient for Action". Article linked in comments. #rstats #statistics https://lnkd.in/g-U9buJu

We're trying out YouTube! In this first video of the "Inside the Blog" series, we go through the article on our website titled, "5 Ways to Help Ensure Success of a Statistical Project". Article linked in comments. #statistics https://lnkd.in/gJDzDkYq

A quick example why statistical significance doesn't tell the whole story and should not be the sole basis for decision making or action. Some main takeaways are: 1. Statistical significance pertains to the existence of a relationship, not the size of it. 2. Pay attention to the magnitude of the relationship. This will get you closer to understanding and interpreting whether or not it matters. 3. Translate the magnitudes into real-world implications of using them for decisions or courses of action. Cost-benefit and what-if scenarios can go a long way. #statistics #decisionmaking

A great new mapping package in R! #rstats

I've been thinking about the idea of "research localization": I think it's often the case that statistical results, especially when marketed to convey broad individual implications as is done in many research articles, are just not all that convincing to people. Irrespective of design and methodological concerns, you can always think of subtleties and nuances as to why it doesn't apply to you, a certain situation, a specific subgroup, etc. Yet these are ultimately supposed to be the units of decision making based on the stated result. It's rarely "the population as a whole" (unless it's for policy or something). It's like there's a lack of proof that a hard outcome has been or would be improved if an individual took action. If we "localize" an analytical/research endeavor to a point that makes the path to action more clear (at the expense of a narrower scope), I think that can actually help lead to more tangible impact of those hard outcomes down the road, albeit iteratively and not all at once, because you are intending to actively prove that value as you go (i.e., this tailored approach objectively worked here, now how can we expand the scope?). The point is to first try to realize tangible impact by being narrow and deep in order to create an actionable product. Once that is working, you can think about what you need to do to be start expanding the breadth, casting a wider net. But you're then building off of something that now has been proven to work. It's going beyond stating that an association exists to showing that it matters. That's the sort of thinking that motivates this article. #statistics #research #datascience https://lnkd.in/ghmcvGDw

The Alliance to Host a Healthcare Data Analytics Workshop for Employers in Wausau: https://lnkd.in/gCQbFkcR

In this June blog post, we run a through a quick example why statistical significance doesn't tell the whole story and should not be the sole basis for decision making or action. Some main takeaways are: 1. Statistical significance pertains to the existence of a relationship, not the size of it 2. Pay attention to the magnitude of the relationship. This will get you closer to understanding and interpreting whether or not it matters. 3. Translate the magnitudes into real-world implications of using them for decisions or courses of action. Cost-benefit and what-if scenarios can go a long way. #statistics #decisionmaking https://lnkd.in/gy4KHmv7

In this June blog post, we run a through a quick example why statistical significance doesn't tell the whole story and should not be the sole basis for decision making or action. Some main takeaways are: 1. Statistical significance pertains to the existence of a relationship, not the size of it 2. Pay attention to the magnitude of the relationship. This will get you closer to understanding and interpreting whether or not it matters. 3. Translate the magnitudes into real-world implications of using them for decisions or courses of action. Cost-benefit and what-if scenarios can go a long way. #statistics #decisionmaking https://lnkd.in/gy4KHmv7

Can confidence intervals be interpreted probabilistically? Yes. But only BEFORE the sample is drawn. The probability is assigned to the sample statistic, NOT the population parameter (despite what the top Google images show https://lnkd.in/gSR3E32Q) For example, suppose we estimate the average age of a population with a 95% confidence interval of 50 to 70 years. We do NOT say that there is a 95% chance that the age is between 50 and 70 years. Our sample was drawn from a population with a fixed (true) average age, and that age is either in our realized interval or not. But if we took another sample, there is a 95% probability that the confidence interval from that new sample would contain the true average age value. It all has to do with the luck of the draw of the sample we end up getting. The intuition here is that because we often assume, especially for averages, that the sample statistics (i.e., quantities computed from the data) follow a normal distribution. So only ~5% of the time we'll draw a sample where the resulting statistic is more than 2 standard deviations from the true average. Then, when we add a margin of error to that estimate (i.e., the lower/upper bounds of the 95% confidence interval, which is just an estimate of 2 standard deviations), we will have added those 2 standard deviations onto either side of the estimate, and thus the resulting interval will only not cross the true mean value if the sample average happened to be in one of those extreme tails, which happens 5% of the time. #statistics

Can confidence intervals be interpreted probabilistically? Yes. But only BEFORE the sample is drawn. The probability is assigned to the sample statistic, NOT the population parameter (despite what the top Google images show https://lnkd.in/gSR3E32Q) For example, suppose we estimate the average age of a population with a 95% confidence interval of 50 to 70 years. We do NOT say that there is a 95% chance that the age is between 50 and 70 years. Our sample was drawn from a population with a fixed (true) average age, and that age is either in our realized interval or not. But if we took another sample, there is a 95% probability that the confidence interval from that new sample would contain the true average age value. It all has to do with the luck of the draw of the sample we end up getting. The intuition here is that because we often assume, especially for averages, that the sample statistics (i.e., quantities computed from the data) follow a normal distribution. So only ~5% of the time we'll draw a sample where the resulting statistic is more than 2 standard deviations from the true average. Then, when we add a margin of error to that estimate (i.e., the lower/upper bounds of the 95% confidence interval, which is just an estimate of 2 standard deviations), we will have added those 2 standard deviations onto either side of the estimate, and thus the resulting interval will only not cross the true mean value if the sample average happened to be in one of those extreme tails, which happens 5% of the time. #statistics