# Homework Help: Combining data from replicated experiments?

1. Jul 27, 2012

### gravenewworld

How do you go about combining data from replicated experiments?

Should you just combine everything into one big set of data and then analyze it, or do you analyze the data from each experiment, and then combine the analyzed data into one set of finalized data? I'm not too sure on what to do, and have been googling for some answers, but there seems to be no clear consensus I can find. Basically all I did was run an experiment where I tested X compound at only 1 concentration for CYP activity by measuring fluorescence. Each experiment I ran I had triplicates. I ran the experiment 3 times. So do I combine the 9 results or analyze the triplicates individually and somehow combine them? I've never had a stats class so am struggling a bit with biostats/data analysis.

2. Jul 27, 2012

### Rooted

Each time you ran the experiment did you run it the same way? I mean was each triplicate ran identically, and each of the 3 times run identically too?

3. Jul 27, 2012

### gravenewworld

Yes and yes.

4. Jul 27, 2012

### Rooted

I'm not familiar with the particular experiment - what kind of output did you get from each run, a single value or a histogram or something?

5. Jul 27, 2012

### gravenewworld

Just a single value for fluorescence. A relative light unit.

6. Jul 27, 2012

### Rooted

If it was me I would combine all 9, I think I've done something similar in the past and worked out that either way of combining the results actually gives the same final results and error, but I would do a mean, and standard error on the mean for the 9 results. The standard error on the mean reduces the error on the mean by a factor of sqrt N, where N is the number of readings.

7. Jul 27, 2012

### gravenewworld

Alright thanks.

8. Jul 27, 2012

### Ygggdrasil

Sometimes it's worth trying different types of analysis to see if they all converge on the same result. For example, you could examine all three sets of triplicates separately. If the variance between the experiments done on different days is similar to the variance within measurements take on the same day, that's good evidence that the experiment is reproducible and you can combine the measurements from the different days (and have a dataset of N=9). However, if you see that the measurements on different days are much less consistent than measurements done on the same day, you may want to use each set of triplicates as an independent measurement (and have a dataset of N=3).

9. Jul 27, 2012

### gravenewworld

So would this require a meta analysis in order to combine all of the data if I treated this as 3 independent measurements? 2 of the experiments seem to be close together while the 3rd one seems to be different. The assay was run from the same kit, the same conditions, just on different days.

10. Jul 28, 2012

### Ygggdrasil

If the experiments were pretty much identical on different days, then you are probably justified in combining the data as a data set with N=9.

If there were potentially important differences between the days (for example if the assay involves cells and you grew up different batches of cells for the experiments on each day), it would make more sense to treat the the data from each day as an independent measurement, and calculate the average and standard deviation from the set of three averages obtained on different days.