Hello Parisa, I don't really know the answer, but if you go to the favourite links page on my website and click on the electronic textbook, it has a definition but I can't see much more detail then this in that chapter, but maybe check out the links that are included I don't have time just now so maybe this is of help. Good luck, Debbie

Deborah Hilton Statistics Online

http://sites.google.com/site/deborahhilton/

Statistical Advisor, How To Compare Survival / Failure Times

Use SURVIVAL ANALYSIS. This chapter describes several tests for comparing survival/failure times in two or more groups. These tests are applicable to censored or uncensored data sets.

SURVIVAL ANALYSIS contains five different (mostly nonparametric) tests for censored data: Gehan's generalized Wilcoxon test, the Cox-Mantel test, the Cox F-test, the log-rank test, and Peto and Peto's generalized Wilcoxon test. A nonparametric test for the comparison of multiple groups is also available.

A data set is said to be 'censored' if some observations are incomplete, but not missing. For example, in industrial reliability testing, a part or product may not fail (break) within the time span covered by a study. However, we do not know how long it will function properly thereafter, and thus, that observation is 'censored.' If your data set contains no censored data points, you can also use the tests for comparing groups or samples discussed in NONPARAMETRICS AND DISTRIBUTIONS.

Dear Debora,

Many thanks for your help.

The page suggested by you looks amazing.

Many thanks for your answer as well.

I think my data is also considered to be censored. I really have to think about it.

Thank you again

If you start to learn how to analyze your survival data, I suggest not to get stuck in a pleiotropy of ancient methods and learn some more modern approach like the Cox proportional hazards model:

P. McCullagh and J. A. Nelder (2000). "Chapter 13: Models for Survival Data". Generalized Linear Models (Second ed.). Boca Raton, Florida: Chapman & Hall/CRC. ISBN 0-412-31760-5. (Second edition 1989; first CRC reprint 1999.)

You can use for instance "bottle" as a possible covariable in such models.

Dear Jochen,

Many thanks for your suggestion. Seems very interesting. I am very new in this. I am going through the method to see how I can use it.

Parisa, first of all you need to know if your data are survival data; then log-rank test is a option. The anova for this problem, probably be incorrect.

Dear Felipe,

My data was survival data over time. The problem is, the test was carried out with group of 20 animals in 7 replicates and not individuals. There was no possibility of doing the test in an individual way. However, I can assure you that the survival of an individual was completely independent from the other.

Parisa.

discusses the possibility of using the Kaplan-Meier , is specifically used for survival analysis

Just a note go keep the overview: Kaplan-Mayer is a way to represent survival data. The Log-Rank-Test is based on Kaplan-Mayer estimates, and the Cox Proportional Hazards is a generalization of all this, including the possibility to consider additional covariables in the models that may be influencing survival.

Dear Carlos and Jochen,

Many thanks for you suggestion.

I have tried the Kaplan-Meier test also in spss. With all the statistical analysis I have used so far, I have significant differences. The problem is, I dont know which one is the most presentable one. My only fear is the fact that in survival study the possibility of an event (0=death and 1=survive) stands for one individual. It means my test was carried out in 140 replicates in my case. However I have 7 biological replicates containing 20 animals. I need to exhibit the most realistic results. I dont know if using survival curves would be still fine when I am sure the survival of an individual animal was independent from other?

What is a "biological replicate" in your opinion? I thought that each individual animal *is* a biological replicate. It might be that a certain group of individuals share a common source of variance, like for instance being kept in the same environment/cage/habitat/time or having the same sex, similar age or things like that. This might be the "bottle" if I understood it correctly. If such variables (cage, sex, age, bottle...) are not correlated with the exprimental variable (treatment or whatever you actually want to analyse) (i.e., not all treated animals are kept in one bottle and all untreated are kept in another bottle) then you can "adjust" the results of the survival analysis by these variables. Well, it might be that the bottle has no influence on the survival, then it is not required to take care about this. But if it has, then the results will be better whe it is appropriately considered. This means: modelling the dependency-structure between different individuals. This seems to be exactly what you are looking for.

Dear Jochen,

Many thanks for your complete reply. That is exactly what do I need to know. There is no influence of the bottles on the animals. The treatment is independent of the bottle as well and it acts really on an individual. So in your opinion, there will be no problem in considering the animals as an individual. Am I right?

Then I can consider the 0 and 1 event and accept any of the results with survival curves I have obtained so far.

Parisa.

sorry,,, but Your question is: considering the group of animals as an individual?

Dear Carlos,

Sorry for the confusion. No. Indeed not. Considering the individual animal as an individual event even though the experiment was carried out with animals in groups. However, the survival of one animal was independent from the other. I have carried out the experiment in 7 replicates containing 20 animals each. considering them as an individual will make me to have 140 replicates of my treatment which reduces the standard error. I was not sure if this will be a possibility.

OK... i now understand something., in reality the information source, always is the experimental unit, in this case the animal.

Now, i continue a bit confused, because, no, you want to show that some condition makes a difference with another . Or would you rather prove that a feature increases or prolongs survival.

So you do have 140 animals = 140 individuals = 140 biological replicates.

Dear Parisa,

I found your question just now. Maybe you still need some advice.

I recommend to use free R for survival analysis. There you can find even examples, which may guide you in your scientific question. Just install R and the library survival and then type:

> library (survival)

> ?lung

To get the Kaplan-Meier estimate of survival depending on sex just type:

> fit1 fit1

To plot the survival curve type type:

> plot(fit1, xlab="Survival Time in Days", ylab="% Survivorship", col=c("blue","red"))

To build a model including interaction:

> coxphfit1 anova(coxphfit1)

Finally, you have to test for the proportional hazards assumption and for linear assumption of coxph model.

Hope the answer is helpful. If you might need more advice let me know or check out my statistics book covering survival analysis and how to write results in a publication.

https://www.smashwords.com/profile/view/altincicek

Best,

Boran

I just found this question. I was wondering  two things: first, how did you determine if the "bottles"  is a variable are correlated with the treatment.  And secondly in case it is correlated how can you adjust  the results of the survival analysis.  I think I have an idea of how to determine the first one. I could run an ANOVA   considering the mean of the survival K-M curves, this comparing the bottles that have the same treatment. If they are not significant differences then I can treat each fly as an indepentent individual. Is that correct? What if there is a difference between the bottles? How can I adjust the results in this case? 

Thank you!

Dear Priscila, 

We solved the problem by performing logistic regression analysis. For survival analysis (overtime) we cant use ANOVA or t-test as your really have to consider the significant difference over time. 

Logistic regression analysis provides you with the possibility to evaluate the significant effect of your treatment on survival over time and at each time point. It is the best when you have a group of animals.