Is it valid to use the mean and standard deviation from a similar, but not exactly the same, study to estimate the sample size for a retrospective study? For example, the reference study was done in patients with diabetes, but the retrospective research will be done in patients who had a stroke.
Using the mean and standard deviation from a similar study to estimate the sample size for a retrospective study can be a reasonable approach, but it is important to consider the differences between the two populations being studied. In your example, patients with diabetes and patients who had a stroke may have different characteristics that could affect the outcome being measured. It is important to ensure that the sample size calculation is based on the expected effect size in the population being studied, rather than just using the mean and standard deviation from a similar study. If there are significant differences between the populations, it may be more appropriate to conduct a pilot study to estimate the effect size in the population of interest.
Hey, thank you for the video, that did help a lot. But I've got a question. How can you do this, when you have an withing-subject-design. My dependent variable is the correct solution of a task (2 conditions: yes/no) and every participant goes through all conditions of the independet variable
That is true if you already know what you want to change. This video is more about exploring what kind of errors there are in the dataset and also, if possible, to identify the exact record and go to the original source to see what the correct input should be.
The information to put in is obtained by; a) read previous publications describing similar studies as yours, and then b) consider what you think you will get in your study and c) decide the reasonable assumptions you need to put in for the sample size calculation. This becomes slightly more difficult if your proposed project is unique and there are no previous publications of similar studies. In that case you simply have to make qualified guesses to come up with the assumptions you need to put in. I hope this reply helps somewhat :)
@@science-network That is great explanation. My project is a newly one. There is no previous publications. So, should I consider effect size as 0.8 to minimize the sample size as I strongly believe that I will reject my null hypothesis? Thank you
@@science-network Thanks for your reply! Similar to how you explained how to obtain the probabilities used in the above method (e.g., assumptions made based on previous publications), it would be great if you could include a similar explanation for how we should determine the probabilities used for the continuous predictor method. Thanks!
Considering interaction between variables in a sample size calculation is complex and most researchers would ignore this when calculating sample size. If you want to dig into this have a look at this: journals.plos.org/plosone/article?id=10.1371/journal.pone.0207745 (sorry but I can't help you further with this).
It is extremely bothersome to me that this series is not being watched by more people. Speaking for myself I am hoping to start my Ph.D. in the coming months and I cannot tell you how informative this video is. I would like to take the opportunity to thank the creator and the academics participating in this video for giving their valuable time and knowledge. I hope with time more and more people realize the importance of this wonderful series.
Thank you. But I have a problem. Somehow, frequencies are not shown in the output for one of my variables. Also, when I am in the frequency analysis window, this variable is in brackets. Do you know what the problem could be?
You don't calculate sample size to achieve a specific R^2. R^2 is relevant when you calculate sample size for multivariate regression such as logistic regression. You may find some guidance about this here: science-network.tv/sample-size-estimation/#Sample_size_calculation_in_multivariate_regression
You have to be CONNECTED to get into these PhD programs. 1,000+ candidates applying for each position. Best if your parents are themselves PhDs currently teaching at a university. Very incestuous.
It might be like that at some places. However, that is not my experience. None of my parents had University or even high school exam. The places where I teach have not been "incestuous" but rather fair. In conclusion don't assume it is bad in all places. If you want to learn about research and perhaps also make it your living go for it! You can read more about prerequisites here: science-network.tv/introduction-to-research/#What_are_the_requirements_for_doing_research
But what if there's no way to fact-check and find what the value "should" be? I have coded impossible values as missing, but I'm not sure where to go from here as far as making my data set clean...
Normally you plan your project so that one column / variable is a unique id for each participant / observation. It would be a good idea to have another file linking up the unique id with the actual real data. You need to consider integrity and protecting data in case individuals (patients?) are involved so the link to real data might be stored in a separate encrypted file having the unique id with the link to how to find the raw real data. The link to real data might be a patient identification number in the electronic chart of a hospital. You are in some trouble if you have no link to real data making it impossible to check suspicious values. Also never throw away hard copies of questionnaires or any other research related material until several years after last publication. So what should you do if you have no possibility to check with the true raw data? Some values are obviously wrong and I support changing them to missing data. However, quite often you find data that are unlikely but not impossible. Are they wrong or correct? There is no way to sort this out without a link to the actual data. My suggestion is to eliminate "impossible" observations coding them as missing. I would also suggest leaving unlikely but possible observations as they are unless you have a very good reason to believe it is an error.
The video about continuous variables is still under construction. For other information around sample size calculation please have a look at: science-network.tv/sample-size-estimation/
