Hello every one, i have a list of genes that i have identified as being associated with anxiety when downregulated and now i would to understand if this genes are also downregulated in other psichiatric diseases. I have downloaded several GEO datasets of several experiments, and i have already performed the mean expression of both control vs target groups but i haven´t found any differences between groups in all the experiments that i have looked.
Could someone suggest me more robust approach?
Do the data or residuals conform to the assumptions of the statistical models?
Do you have enough data to reliably answer the above question?
Significance is not a statement of truth or importance. It is only a probability of the null hypothesis being true given the observed data and assumptions. You cannot prove the null-hypothesis, though we sometimes mistakenly try this. A common situation where we treat non-significance as proof of the null hypothesis is in testing for the assumption of normality.
The p-value is also an estimate. With the right data you could construct a 95% confidence interval about the p-value. Generally we do not have enough data to do this.
Hi, Dr. Carvalho,
An interesting research application! In the event that no legacy statistical method provides you with a satisfactory solution , perhaps the following is worth a look.
The modern machine-learning-based optimal data analysis (ODA) paradigm employs mathematical programming to identify maximum-accuracy solutions. This paradigm compares entire distributions, rather than comparing point estimates such as means or medians. ODA requires no assumptions regarding data or residuals, provides exact p-values, explicitly identifies the most accurate model possible--with effect size normed against chance--for your data and hypothesis, and conducts many types of cross-generalizability analyses.
Here is a brief open-access article which reviews the ontogenesis and etiology of the ODA paradigm:
https://odajournal.com/2017/04/18/what-is-optimal-data-analysis
Many open-access articles are available at https://odajournal.com
The Publications Tab of the eJournal provides a sorted list of many similar articles published in pay-for-view journals.
If I understand correctly, you have multiple data sets, each with at least one outcome, and each with a treatment and control group? In general, the simplest and most parsimonious analysis is preferred as it is typically easiest to replicate in future studies. If you have two groups, then a simple t-test would probably be the best way to compare the means. As mentioned above, statistical significance is biased by sample size and doesn't tell you about practical effects. So, another thing to consider is calculating and comparing effect sizes (Hedge's g or Cohen's d) across all the experiments you have looked at regardless of the statistical significance. Because they are standardized mean differences, you can directly compare them across your experiments regardless of the different outcomes you use. Hope that helps, best of luck.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis