In statistical analysis, can i use a file with the missing value or not?
i want to do normality test so could i find what test is used for my dataset.
If the analytical software supports missing values, yes you can use files with missing data. SAS System, in example, represents missing values as a dot (.). Be carefull, some softwares does not supports missing values and empty cells are treated as zero (0). Small samples (
Hello Lakhan -
Do not be too concerned with normality. The central limit theorem usually handles most problems with that, say if you wanted a confidence interval for a mean. You can also consider Chebyshev's Inequality.
Hypothesis tests are notoriously misused and the users misled.
As for missing data, if they are missing at random, that may be ok. Otherwise they bias results.
Sample size/variance also matter, and measurement error matters.
Cheers - Jim
Thanks, Dr. James R Knaub and Dr. Marcelo Corrêa Alves. Can you provide some literature of chebyshev's inequality test and Shapiro-Wilk test.
my sample size for the test is small. If i do normality test with the help of SPSS statistical software then i can use missing value or not?
In statistical analysis, can i use a file with the missing value or not?
Suggesting if you have small precious sample size, for those rows / respondent records with missing value, try to replace with mean value calculated based on that column field. If you have large sample size, you can exclude those rows with missing value(s).
Since you already have a small sample size, it is not advisable to discard further data due to missing values. Replacing with mean value as suggested by HAN PING FUNG will be the better way to go.
1) Chebyshev is found in many places, in various forms, in texts and on the internet. I looked and here is one I just found:
https://www.thoughtco.com/chebyshevs-inequality-3126547
2) With regard to missing data, as I previously stated, imputing mean values will bias results if the nonresponses were not at random. This could be especially harmful for small samples. A high nonresponse rate is highly undesirable.
If you have auxiliary data to impute by regression, that could be good.
Perhaps your data set is large enough to group into two (or more) strata, based on a population characteristic, and you can see if means are substantially different between strata. Then, the least you could do would be to treat those strata separately.
If you have any reason to think the missing data are not missing at random, you may need further study.
At any rate, you should be aware that just replacing a missing value by the mean of the responses can bias results, and those data cannot be used in estimating variance as they artificially reduce variance, because 'formulas' would assume the imputed data to be real data, which they are not.
So, beware that you may not have reliable results.
Dear Lakhan
In the case of random missing data you can perform your data analysis as unequal replicated data, but in some cases like Randomized complete Block Design ( RCBD ) you need to make the estimation for the missing data before doing the analysis.
Good Luck
Dear Lakhan
Yes you can. This is largely dependent on the type of analysis you intend to employ. If the analysis is related to a survey, it can be regarded as a non-response and the researcher should be aware of the concept of non-response bias. If the analysis is not a survey, dependent on the quality and distribution of the data you may decide to use the "Mean Replacement method" to replace missing data (this method is not recommended in most cases). Otherwise if you want to continue analysis and disregard the missing value then you should use a statistical software that treat missing data as missing and not as zero.
Good Luck
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis