Use the thermodynamic scale for Temperature (Kelvin).

You have outcomes ranging from 0 to 50% but the outcomes could be from 0 to 100. I think the simplest option might be the empirical logistic transformation:

elogit = ln ( (P + c) / (1-P + c) )

where P is percentage/100 and c is a small constant such as 1/1000.

@ Jochen Wilhelm thank you sir for the answer but I dont think so if I should use thermodynamic scale for conversion, because the result refers to the number of eggs and number of the other phase not the treatment

@Thom S Baguley thank you sir for the answer, I have used empirical logistic transformation but the result is (-) minus

Afuar, oh yes, you are right! I apologize my stupid answer. I got the question wrong. Sorry.

Your response variable is - if I now understood correctly - binomial. That means you have some number some number of eggs (n) and from those eggs you get a number k of "surviving" larvae (or pupae or adults). So the number k can be modelled as binomial-(n,k) distributed. You can use a binomial model ("generalized linear model or the binomial family with logit link") to model your response and make inferences on the experimental factors (e.g. temperature).

If you don't get eggs at 40°C then this temperature is out of the range you can investigate. It is notable that there are no eggs for this temperature, but it makes no sense to try investigating mortality rates for this case. So I would exclude this from the further analysis (i.e. the binomial model).

Regarding your experimental design I have a problem:

You state that you want to analyze the effect of the constancy of the temperature (varying around 27°C). This would require to have a group that is grown at a constant temperature of 27°C, but I did not see such a group.

The information from higher but constant temperatures don't help you finding out what temperature variations during development do. You might "only" use this to find the optimum temperature (at what the mortality is minimal), although having only 3 different temperature values is a bit few to estimate this optimum.

I hope I wasn't too wrong this time.

@Jochen Wilhelm Thanks for the answer again, I consider your answer and I should disscuss it with my supervisor. Thank you sir.

Because the research tools such as incubator or growth chamber (I used baby incubator) in my departement not too much as I want so I just researched 'only' 3 constant temperatures, and the room temperature (27oC) is comparison if the farmers or people who works in the tobacco industry do not control temperature for lower the population of Lasioderma serricorne.

The following is simply brain storming to supplement the second answer by @Jochen Wilhelm (above)

Although you can discuss it in your results, the 100% 40C data is not suitable for inclusion in the test. The difference is obvious. Then, if every treatment has a sufficiently large sample size you might be able to use an ANOVA, but a non parametric Friedman's test would be much better, especially in light of the zero's.  Maybe the old Practical Statistics Simply Explained, by Russel Langley would be an easy reference, however Soakal and Rholf Biometry also has it.  With a Bonferroni correction, you could the use pair by pair multiple comparisons using a non parametric test like Wilcoxon's test.  Some graphical approach might be a good idea too. At least include (bar?) graphs, which could show the 40C result. You might try the graphics with the results converted in various ways, but transformations are a mess to use in the actual analysis. 

add one in digit or observation and take square root of the value  and out can be used for calculation

You could consider a non-parametric data analysis (Kruskal-Wallis and Dunn tests) like we have done for zero (and higher) mound counts, in which case no transformation is likely necessary...

Article Distribution and density of termite mounds in the northern K...

Hi

A few questions to try to understand your data......

What's your n? What is your experimental unit? This is relevant either you go for ANOVA or Kruskall-Wallis test?

If you are following a cohort of insects, you may use a Kaplan-Meyer analysis to describe the mortality and compare every curve you have..................

Your data (number of eggs, larvae, pupae and adults) is not normal, so be cautious before going for an ANOVA unless your n is large enough go for a parametric test, assuming normality and variance requirements are meet. The KW test is the obvious choice.  

Notice that KW and Friedmans tests will compare different things.............

I agree with Mukesh Singh, square root of x+1 is a good option for data with many zeros

if you have percentage data, you may use arcsen transformation before ANOVA and here again the sample size is very relevant because the df of SSof the Error

Cheers

Alf

Dear Dr. Aufar Akaar, The best for statistical calculation when some of the numeric data are zero (= nile = 0), is to trasnform all data iinto  log x+ 1. Then you use any statistical model (ANOVA, non-parametric model, F test ...). Sincerely yours. [email protected]