One way ANOVA or two way ANOVA - (Jan/24/2007 )

Hello everyone,

I'd like some advice on statistical analysis for my experiment. Basically I am transfecting my cells with different plasmids and measure results by % GFP+ cells on the FACS. I've done all transfections in triplicate (ie I've made a transfection mastermix for each plasmid and divided it into three wells) to account for measurement variation and then repeated the experiment the following week to account for experimental variation.

So now I'm not sure which ANOVA test to use. A statistician (clinical statistician and didn't quite seem to understand my experiment) said two factor ANOVA with replication (as in Excel), however from my research into this subject I think it should be a one way ANOVA with repeated measures (average of triplicate in first experiment should be paired with average of triplicate of same plasmid in second experiment).

Any comments or help would be most welcome.

Regards

LeserattePD

-LeserattePD-

QUOTE (LeserattePD @ Jan 24 2007, 04:15 PM)
Hello everyone,

I'd like some advice on statistical analysis for my experiment. Basically I am transfecting my cells with different plasmids and measure results by % GFP+ cells on the FACS. I've done all transfections in triplicate (ie I've made a transfection mastermix for each plasmid and divided it into three wells) to account for measurement variation and then repeated the experiment the following week to account for experimental variation.

So now I'm not sure which ANOVA test to use. A statistician (clinical statistician and didn't quite seem to understand my experiment) said two factor ANOVA with replication (as in Excel), however from my research into this subject I think it should be a one way ANOVA with repeated measures (average of triplicate in first experiment should be paired with average of triplicate of same plasmid in second experiment).

Any comments or help would be most welcome.

Regards

LeserattePD

For counts/numbers I'd chose the chi-squared test (although I'm not sure which type of data and amount of data points you have). It is not a repeated experiment, as the second one uses different samples. Th experimental design is only repeated if you use the same samples (from 1st experiment) for measuerments a second, third time, etc .

-hobglobin-

[/quote]
For counts/numbers I'd chose the chi-squared test (although I'm not sure which type of data and amount of data points you have). It is not a repeated experiment, as the second one uses different samples. Th experimental design is only repeated if you use the same samples (from 1st experiment) for measuerments a second, third time, etc .
[/quote]

I think you misunderstood - I've done excactly the same experiment in the repeat, including using the same samples ie the same preparation of plasmid, the same conc, same transfection protocol, same cells just two passages older. According to the GraphPad Prism Statistics Guide, one should choose "repeated measures" if "You run a lab experiment several times (twice in my case) each time with a control and several treated preparations handled in parallel."

-LeserattePD-

If I do this , i will choose one-way ANOVA with repeated.
Different plasmid and same other conditions can be seem as one factor at different levels , so One way ANOVA maybe useful.
Chi 2 test maybe useful for the contingency table data.
Of course , if you use different cells or other materials , you also need to confirm these factor's influence on your results. So at this condition , the complex ANOVA will be usable.

-boyang-

I will apply one-way ANOVA because it seems there is no variation, you made only one treatment but triplet replication.

Compare to this research of mine, which used two-way ANOVA:

Groups:
1. Administration with extract protein fraction
Subgroup (dose-based): 100, 200, 300 mg/kg bw
2. Administration with extract lipid fraction
Subgroup (dose-based): 100, 200, 300 mg/kg bw
3. Control

Each group/subgroup are observed on day-0, 34, and 67
Each group/subgroup consists of 5 replicates.

-Miss L-

QUOTE (Miss L @ Jan 25 2007, 09:38 AM)
I will apply one-way ANOVA because it seems there is no variation, you made only one treatment but triplet replication.

Compare to this research of mine, which used two-way ANOVA:

Groups:
1. Administration with extract protein fraction
Subgroup (dose-based): 100, 200, 300 mg/kg bw
2. Administration with extract lipid fraction
Subgroup (dose-based): 100, 200, 300 mg/kg bw
3. Control

Each group/subgroup are observed on day-0, 34, and 67
Each group/subgroup consists of 5 replicates.

Good god - I don't even want to think about the statistics for that!

-LeserattePD-

Hi LeserattePD,

if you need a one or two way ANOVA depends on the question you want to have answered. I suppose, you like to know if different plasmids lead to different FACS readings? Than you're interested only in one factor that is different between the groups - so you need a one-way ANOVA. If you want to know, if not only the plasmid but also the day of preparation has an impact and if this impact may interact with the effect of the different plasmids, choose two-way ANOVA. But are you interested in that?
Regarding repeated measurements - a good rationale is to compare the readings of the different days - if they are pretty similar, you can just combine all 6 measurements (if you use prism, use 6 replicates for every y). If you have large differences between the days than you can apply one way ANOVA for repeated measurements (which, in fact is a two way ANOVA, incorporating the time as second factor).
So, you were right AND the statician was right - cool!

Hope that helps,

Krümel

-krümelmonster-

QUOTE (krümelmonster @ Jan 30 2007, 08:43 AM)
Hi LeserattePD,

if you need a one or two way ANOVA depends on the question you want to have answered. I suppose, you like to know if different plasmids lead to different FACS readings? Than you're interested only in one factor that is different between the groups - so you need a one-way ANOVA. If you want to know, if not only the plasmid but also the day of preparation has an impact and if this impact may interact with the effect of the different plasmids, choose two-way ANOVA. But are you interested in that?
Regarding repeated measurements - a good rationale is to compare the readings of the different days - if they are pretty similar, you can just combine all 6 measurements (if you use prism, use 6 replicates for every y). If you have large differences between the days than you can apply one way ANOVA for repeated measurements (which, in fact is a two way ANOVA, incorporating the time as second factor).
So, you were right AND the statician was right - cool!

Hope that helps,

Krümel

Liebes Kruemelmonster,

Thanks a lot for your help. I was really interested in the differences between the plasmids, but also to see if the differences between experiments were significant. After all if I can't get reproducible results I can't use them to claim that something is true.

So I have decided that in future I'll need to do the two way ANOVA to check for significant differences between the experiments (in the case dicussed above it was F=2.2 and F crit 4.4, p=0.156 so no significant differences) and if there is no significance then I use the one way ANOVA with repeated measures and Bonferroni tests to get a nice presentable graph, where I can state that it is from repeated experiments and show p values.

I think one of the things that confused me was the fact the I sort of thought that the repeated measures one way ANOVA was the same as the two way ANOVA, nice to hear that that was right.