I have observed that the regression analysis of log conc vs probit (%mortality) in ms-excel can provide you the values for anova, r^2, se, intercept and slope but the values for intercept and slope are entirely different in spss output for the same date in mortality.
Spss takes mean mortality and automatically transforms it into probit values but the lc50calculated from both gives difference in points.
LCL and ULC in excel are only provided for slope not specifically for Lc50 how to calculate it?
Although in spss x2 and LCL-UCL are provided.
Inshort which one provided more accurate data spss's probit analysis or msxcel's regression analysis of probit values
Hi Savy. My first advise is to consider SPSS results as the more accurate ones. The reasons behind the {intercept,slopes} may be explained by the specific "relevant point" (of the explanatory variables) at which MS Excel computes the marginal effects (besides, one could be providing the actual parameter estimates instead of the marginal effects). Before I can provide more help, let me understand the regression exercise you are doing. First, you wrote "...analysis of log conc vs. probit (%mortality)..." What do you mean with "log conc", the log concentrated likelihood? What do you mean with "%mortality", the probability of the event "not to be alive"? You also wrote "... but the values for intercept and slope are entirely different in spss output for the same date in mortality" What do you mean with "the same date in mortality"? Is it the date when the event "not to be alive" becomes a realization? Is it then some kind of survival analysis? Please define Lc50, LCL, ULC & UCL before using them (although I think ULC is a typo). Finally, I recommend you to program the whole Probit estimation routine in Matlab. Mathworks has a 30-day free trial downloadable version with many toolboxes nowadays (try the "data analysis" trial version). I bet you can take a good econometrics text like Professor Greene's and program it while learning how to load data, plot it, estimate a log(odds) simple regression and then define the (minus) log(likelihood) function to be minimized with the available optimization tools therein. Best regards!
Carlos R. Barrera-Chaupis
Dear sir,
1. can you please confirm the authenticity to use the probit analysis using this method in the attached website.
(both are same)
https://probitanalysis.wordpress.com/
https://drive.google.com/file/d/0B5gMWH27EEODX0JzYzU4RGNlNTg/view
2. I have replied to your questions below:
>First, you wrote "...analysis of log conc vs. probit (%mortality)..."
yes
> What do you mean with "log conc", the log concentrated likelihood?
the main concentration was converted into log10 values
>What do you mean with "%mortality", the probability of the event "not to be alive"?
yes, the mean mortality data (which shows the no. of larvae dead in the test) was converted into the percentage and that percentage was transformed into probit value.
>You also wrote "... but the values for intercept and slope are entirely different in spss output for the same date in mortality" What do you mean with "the same date in mortality"?
I meant that the "data" (mistypes as date)
It gave me entirely different values as:
the data I have entered in the spss was not in percentage but only the mean of the 3 replicates.(spps was able to convert the data automatically into the probit values and conc. into log10 conc. to calculate the slope and intercept)
>if I enter data in % in spss then also the values for the intercept and slope are different than that if in excel.
>in ms excel I did a regression analysis of the transformed mortality (%) in probit values (y-axis) against log conc. (x-axis) which gave me different values for slope and intercept
now these values can give me the x - value which differed in points (at y=5).
but my question here is then which method to go with.
> Is it the date when the event "not to be alive" becomes a realization? Is it then some kind of survival analysis?
sorry it was data (mistyped)
> Please define Lc50, LCL, ULC & UCL before using them (although I think ULC is a typo).
Lethal concentration 50 (50% probability of dead larvae in the sample), LCL&UCL-lower and upper confidence limits at 95% level of significance.
Finally, I recommend you to program the whole Probit estimation routine in Matlab. Mathworks has a 30-day free trial downloadable version with many toolboxes nowadays (try the "data analysis" trial version). I bet you can take a good econometrics text like Professor Greene's and program it while learning how to load data, plot it, estimate a log(odds) simple regression and then define the (minus) log(likelihood) function to be minimized with the available optimization tools therein. Best regards!
thank you very much for your advice and wishes. I will try matlab also thanks for the suggestion.
kind regards
savy
With a lot of happiness, I confirmed the authenticity of using probit/logit analysis as a method to compute the median lethal dose (LD50) aka the median lethal concentration (LC50). This method was developed by Finney (1952) "Probit Analysis", Cambridege U Press. I was later discussed by a study made by Robertson et al. (2007) - but the exact reference for Robertson et al. is missing in the link https://probitanalysis.wordpress.com/
However, I should say that this method can be drastically improved by using a "beta regression", where the dependent variable is a proportion or a probability in the interval (0,1). As we can also confirm, "probit/logit regressions" are specially tailored for a dichotomous dependent variable that takes only two posible values, one or zero, {0,1}, instead. See the corresponding chapter of William Greene, Econometric Analysis (I prefer this book's editions of the 90s, but I am sure the recent editions will also cover this material well).I will really appreciate Dr. Panamkuttiyiel's providing the exact reference for Robertson et al. (2007)'s study, because I bet this recent reference to the LD50/LC50 method mentions this issue. Best regards.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology