I want to collect (all) melioidosis cases between years 1995 till 2014 to study patient's data variables such as demography, type of melioidosis, predisposing factors and clinical outcomes. I'll do association analysis using regression and so on.
1. so for this project: the study design is Retrospective?
2. Shall I calculate sample size (remember I'm willing to take all cases within defined year range
3. Or some how, shall I use power sample size?
It is not the issue of the design, retro or pro. but the study hypotheses or the research questions. If you will use inferential statistics for example to compare 2 groups, or 2 time periods, you may use the ttest. etc. thus you can use G. power program which is available free online to calculate your sample size based on ttest
You can use PS or G power software to calculate appropriate Sample size.
It is not the issue of the design, retro or pro. but the study hypotheses or the research questions. If you will use inferential statistics for example to compare 2 groups, or 2 time periods, you may use the ttest. etc. thus you can use G. power program which is available free online to calculate your sample size based on ttest
I assume that there are going to be a large number of problems (unexplained variability). You will have to ignore problems with how the data were collected, and hope that you will find a signal amongst all the noise in the data.
I have no idea how common these cases are... So get the first 50, and run the analysis. Gather another 50 and rerun the analysis. Repeat a couple more times, and then look carefully at the output. What does the addition of 50 more patients do to the regression output? Given that these patients are not all from the same population, the regression model could become better or worse with each addition of 50 patients.
You could help this project by having a clear question or hypothesis. The "can I find a correlation" makes the process more difficult. It would also help if you can clearly identify the population being studied. Is this the entire human race everywhere that the pathogen is found, or just everyone at the local hospital? You will also have to figure out what to do with missing data: imputation or deletion? How will you identify and handle errors (errors in measurement, recording data, entering data into computer, misdiagnosis).
I would also factor in how easy it is to get the data. If a key word search gets me all 4875 cases, then I would use all the cases. If it required hours of searching through old archives for each record, I might opt for something much less inclusive.
I tend to advocate a sample size as large as possible. However, some consideration needs to be made for ethical issues and the difficulty or hazard present in gathering another sample. Furthermore, there is a law of diminishing returns: the cost of adding a new sample versus the increase in accuracy/power from adding another sample. So what size effect are you willing to mistakenly fail to reject the null hypothesis? In my research on insects, the ethical issues are less important, but the other issues are all unknowns. So I try to gather as much data within my budget as I can.
I agree with Professor Ahmad that retrospective is not an issue. All data is retrospective at the time the analysis is performed. Many studies are multi-year, and may involve different scientists over the course of data collection.
The issue is how much is the sample, so you do power calculation and if you get the mimimum number in your available data, then you are fine.
To do power analysis to estimate your sample size, you have to write your hypothesis, and based on that you decide what statistical test you will use. It should be one of the inferential statistics. so you need to determine the following: alpha {standard to be .05}, power [standard to be .80], effect size {small, moderate, or large, each test has its own value, you can find these values in the net}. Then download free programs to calculate the sample size such as G. power.
The issue is how much is the sample, so you do power calculation and if you get the mimimum number in your available data, then you are fine.
To do power analysis to estimate your sample size, you have to write your hypothesis, and based on that you decide what statistical test you will use. It should be one of the inferential statistics. so you need to determine the following: alpha {standard to be .05}, power [standard to be .80], effect size {small, moderate, or large, each test has its own value, you can find these values in the net}. Then download free programs to calculate the sample size such as G. power.
The issue is how much is the sample, so you do power calculation and if you get the mimimum number in your available data, then you are fine.
To do power analysis to estimate your sample size, you have to write your hypothesis, and based on that you decide what statistical test you will use. It should be one of the inferential statistics. so you need to determine the following: alpha {standard to be .05}, power [standard to be .80], effect size {small, moderate, or large, each test has its own value, you can find these values in the net}. Then download free programs to calculate the sample size such as G. power.
Regardless if it is retrospective, it is important to know how much is the needed sample, so you do power calculation and if you get the mimimum number in your available data, then you are fine.
To do power analysis to estimate your sample size, you have to write your hypothesis, and based on that you decide what statistical test you will use. It should be one of the inferential statistics. so you need to determine the following: alpha {standard to be .05}, power [standard to be .80], effect size {small, moderate, or large, each test has its own value, you can find these values in the net}. Then download free programs to calculate the sample size such as G. power.
Muayyad Ahmad - can you give an illustration on your point please
Regards
Sample size calculation seems to be irrelevant in this case. AbdelRahman is taking all cases in the targeted years. This is still a sample as I would assume that some cases are not reported or misdiagnosed. Some patients may have gone outside the study area, or (maybe) died before they could be properly diagnosed. One must also consider if the study is intended to apply to a broader population, or only to the local area. However, your sample size is fixed and probably closer to being a census than a sample.
The suggested software is G*Power (http://gpower.hhu.de/). It is an interesting and useful program. However, like all tools it has its limitations.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size