Statistics question.. help if you know

Hi guys :slight_smile:

Dont know mutch about statistics and i thought some of you might be able to help.

I am doing research ELISA and the kit manufacturer says on the manual that i should do 4 parameter logistic analysis of the results i get. I am using GraphPad Prism 4 for the statistics and i cant find anywhere this type of analysis - it simply isnt there.

I know it belongs in the non linear category (i think) but its not in the drop down menu of available analysis. Does it have any other name (synonym) or is there a specific equation that i can fill in in custom analysis?

I can also use SPSS for the data analysis, but i am not sure whether that choice will be available plus i will have to learn to use the program from scratch.

huh?

I’ve only ever used SPSS for analyses, but i don’t know what a 4 parameter logistic anlysis is either…

if you do switch to SPSS though and need some basic starter help I may be able to help you out a bit. That was what I used for all my psych research studies in college.

Hi Hemi

Logistic regression analysis is a fairly advanced statistical technique, designed to analyse the relationships amongst a mass of unconnected binary variables and highlight the most important that might be individually contributing to the end-result.

Basically, if you have this significant association between the mixture of variables and an outcome, logistic regression allows you to withdraw one variable at a time from the mixture and observe whether the level of significance suddenly drops away (ie p goes from .001 to .9). If that does happen, the inference can be made that such a variable is contributing substantially to the outcome. Conversely, if withdrawing a variable doesn’t alter the significance, then it’s possible that variable isn’t involved in the real situation. IIRC the history of the approach is that it was developed during the research into the association between smoking (only one amongst the myriad of other lifestyle factors that could have had an influence) and lung cancer.

I have used such an analysis before in a program called StatView, and it’s very good as long as you enter the data correctly. However I would definitely advise you to consult a statistician for some help; it’s not a standard analysis that you’d expect to find in a stats and curve-fitting program like Prism and I’m surprised you’re advised by the kit manufacturer to do that with simple ELISA results.

If I could ask, what are you trying to measure with this ELISA, and what are the four binary variables (i.e. the +/- parameters)?

It sounds all greek to me :stuck_out_tongue:

The data types i am using are simple and straight forward, let me explain. I get some optical density values from samples of known concentration (standard samples provided ny the manufacturer) and from that i interpolate to find the values of samples of unknown concentration (cytokines in blood/serum etc). Up to now i was using linear regression for getting the values (which i thought was logical to do since the standard specimens have double the concentration consequently ie 200 pg/ml, 400pg/ml - so i expect the densities to be double too which they are). But according to the manufacturer this is not the correct way to process the data.

Let me quote from the kit’s manual:

CALCULATION OF RESULTS
Average the duplicate readings for each standard, control, and sample and subtract the average zero standard optical density.
Create a standard curve by reducing the data using computer software capable of generating a four parameter logistic (4-PL) curve-fit. As an alternative, construct a standard curve by plotting the mean absorbance for each standard on the y-axis against the concentration on the x-axis
and draw a best fit curve through the points on the graph. The data may be linearized by plotting the log of the OPN concentrations versus the log of the O.D. and the best fit line can be determined by regression analysis. This procedure will produce an adequate but less precise fit of the data.

Maybe i dont get it at all… the thing started when i got some results ready from another lab and when i redid the analysis my way the data dindt match - the lab operator told that was because he was doing it with the 4 parameter analysis and not with linear regression as me.

I have no idea


Hehe… OK I see what it is now. Maybe I assumed too much - all you need is an accurate curve-fit for the standards. An ELISA standard curve is basically a sigmoidal shape which plotting log:log will approximate, but it will be inaccurate somewhere along the way.

It depends how accurately you want to measure these things; how far out are the different values with your log:log plots and the fancy ones?

BTW the Prism website has a page for curve-fitting:

http://www.graphpad.com/prism/curvestandard.htm

… and a PDF you can download which explains how to use the program to fit a sigmoidal ELISA curve (see bottom of image for where to click on real site):


Thanx a lot mate! I have some reading to do, will get back to you with more questions if needed!!

Airy, as always you saved the day :stuck_out_tongue:

Yip - PM or whatever…

Indeed, a classic by Air. May use that in a talk next week :slight_smile:

God i hate statistics - my brain is fried. I think its high time i updated my sig…

btw Henky you are the best …

some stats help for ya :wink:

cant open video links with the deer park 64bit browser - god it sucks

imkidd57 i finally got it, your input was crucial in helping me do it. Thanx again - and lets not forget Airy’s contri in this whole thing too :stuck_out_tongue:

Wohoo


Excellent :flower: - let me know if you need any more input…

Indeed: his incisive analysis is completely acknowledged :iagree:.

If anyone can come up with a cytokine joke, he’s our man ;).

Close…

Actually Hemi, I worked 15 years with this stuff. Guess what…I can´t remember too much! Terrible really…no, really not.
The manufacturer is usually able to provide a guideline that can get you there…helpful info anyway. Ok…we also used to make our own standards (wherever possible), but in the end ALWAYS compare your results with the clinical findings. I cannot stress this enough…many interns, and more than enough experienced docs, continued to look for absolute numbers when they didn´t really exist.
Find a good lab tech who can explain (they must exist…that´s what they are trained for!..was my job once).