Dale said:I fundamentally disagree with the Q test and similar outlier rejection techniques. I tend to use graphical methods to visually assess possible outliers. For any possible outlier I look at the data and see if there was a transcription error or a recorded experimental "glitch" or flaw. If there is a transcription error then I correct it, and if there was a recorded "glitch" then I reject it. I never reject data simply based on distributional assumptions.
In particular, the Q test assumes normality. An outlier can tell you that your data is not normal. If you assume normality anyway and reject the outlier then you are ignoring important information telling you that your assumption is wrong.
That said, the Q test is only intended for small numbers of observations. 180 is too many. You should pick a different test (or use a different approach alltogether).
When number of sample is small , we can use Q - test to select outliers data. If we find that there is still outlier data , we can use Q-test to confirm reject this data again.Dale said:I use something similar in my graphical methods. One of the graphical methods I use is a box and whisker plot which uses this calculation to determine if any data points should be plotted individually. I can then look at those individually plotted points and see if there is a transcription error or an experimental reason to reject the data. So I do use that method, not to directly reject data, but to spot data that I should look at in more detail.
No. The Q test should never be used more than once and is only used to reject at most one observation.Another said:When number of sample is small , we can use Q - test to select outliers data. If we find that there is still outlier data , we can use Q-test to confirm reject this data again.
You can use Box Plots as often as you like, but I would not reject any data based only on the Box Plot itself. All that should do is identify data points to investigate. If you investigate and find that there is no transcription error and no experimental problem, then you should keep the outlier.Another said:When number of sample is large , I want to use Box Plot method to select outliers data.
If I find that there is still outlier data , I can use Box Plot method again ?
The Q-test is a statistical tool used to determine the presence of outliers in a data set. It helps to identify values that are significantly different from the rest of the data and may impact the overall analysis or conclusions drawn from the data.
The Q critical value is calculated using a formula that takes into account the sample size, the confidence level, and the degrees of freedom. The formula is Q_crit = (n-1)/(4n), where n is the sample size. This value can also be found in statistical tables or calculated using statistical software.
The Q critical value represents the maximum value that an outlier can have before it is considered significant at a given confidence level. It is used to compare the Q calculated value (from the data) to determine if the outlier is significant or not.
A confidence level of 95% is commonly used for the Q-test because it is a standard level of confidence used in statistical analysis. It provides a balance between being too strict (higher confidence level) and potentially missing important outliers, and being too lenient (lower confidence level) and falsely identifying outliers.
Yes, there are some limitations to using the Q-test. It assumes that the data follows a normal distribution and is not suitable for data sets with small sample sizes (less than 10). Additionally, it can only detect one outlier at a time, so multiple outliers may require a different approach for identification.