# Just need a little more help here...

#### David S

##### New Member
I posted about this issue a couple of weeks back, but unfortunately none of the suggestions, while helpful, really hit the nail on the head. My proposed experiment is described below:

I am investigating the germination of Berberis wilsonii seeds under varied conditions. For this particular part of the work my hypothesis is that there will be a significant association between the length of time the seeds are cold stratified and the germination rate. I have 6 petri-dishes, each with 50 seeds in. one is control, the others are subjected to 2wks, 4wks, 8wks, 12wks and 16wks refrigeration before being germinated...

So, the independant variable here is continuous ratio, the dependant variable discrete count.

I was pointed towards both Pearsons & Spearmans correlations, but as far as I can see, neither will fit my data type. Logistic regression was also suggested, but I can't see how this would fit either.

Now, if I simply divide the total germination figure for each petri dish by 50, I get a figure between 0 & 1 describing the probability of the event occurring; this would be continuous. Now I have both dependant and independant variables continuous. So my question is this: what test should I use to investigate the level of significance compared to the control for each of my five results? How do I work out whether the length of time the seeds are stratified for is significant compared to the control?

All help, as ever, much appreciated.

#### Miner

##### TS Contributor
Have you considered a 1-way ANOVA followed by Dunnett's post-hoc test for multiple comparisons to a control? I assume your control is 0 wks refrigeration. This is one treatment variable with 6 levels. The only drawback is that you will need replicates (i.e., more than one petri dish per level).

#### noetsi

##### Fortran must die
You could look at poisson regression if your DV is a count.

#### David S

##### New Member
Wow. thats it, thanks!
Have you considered a 1-way ANOVA followed by Dunnett's post-hoc test for multiple comparisons to a control? I assume your control is 0 wks refrigeration. This is one treatment variable with 6 levels. The only drawback is that you will need replicates (i.e., more than one petri dish per level).
That sounds ideal. I'm not a statistician; i'm just finishing an assignment for my BSc Arboriculture-I have never come across Dunnetts test-it sounds to be exactly what i'm looking for. Thank you so much for your help with this. I've had the rest of the work written up for a fortnight, i've just been struggling with this to finish it. I'm very grateful!

#### David S

##### New Member
You could look at poisson regression if your DV is a count.
This also sounds good. Amazing. Its like buses-you wait for ages for one to turn up, then they all arrive at once! Cheers guys, Your help has been invaluable and I appreciate you both taking the time to assist me with this.

All the best!

#### noetsi

##### Fortran must die
For someone who is not a statistician and an undergraduate you chose a tough thing to do

#### David S

##### New Member
For someone who is not a statistician and an undergraduate you chose a tough thing to do
Only because of my ignorance!

#### noetsi

##### Fortran must die
lol it usually works that way. After years of studying such things I have gotten afraid to try.

It would be worthwhile IMHO if you are going to do such things to get a good book on design of experiment. This is not statistics per se but how you set up your design and gather the data. Which in my experience is about 90 percent of the work and where things can go badly wrong.

Also you might want to consider the sage monograph series. They are short and a good introduction to nearly anything in statistics. They have one for pretty much any topic you are interested in.

#### fed2

##### Member
you know i was thinking about this a bit. I think you need to use 'germination indices', you may have seen them around, and if not you are likely to in your field. I have seen them in passing in bacterial growth, basically the idea is that you compute various quantities baed on germination curves. I think this is the prevailing method in agricultural studies. You probably just are not doing the logit right, that is also probably a good option,.

#### David S

##### New Member
lol it usually works that way. After years of studying such things I have gotten afraid to try.

It would be worthwhile IMHO if you are going to do such things to get a good book on design of experiment. This is not statistics per se but how you set up your design and gather the data. Which in my experience is about 90 percent of the work and where things can go badly wrong.

Also you might want to consider the sage monograph series. They are short and a good introduction to nearly anything in statistics. They have one for pretty much any topic you are interested in.
Sounds like a good idea, evidently it's something I could do with mugging up on. I'll see what I can find at my University library in Newcastle. I'll have a look at the series you suggest as well.

#### David S

##### New Member
you know i was thinking about this a bit. I think you need to use 'germination indices', you may have seen them around, and if not you are likely to in your field. I have seen them in passing in bacterial growth, basically the idea is that you compute various quantities baed on germination curves. I think this is the prevailing method in agricultural studies. You probably just are not doing the logit right, that is also probably a good option,.
You're probably right about the logit, but I just didn't really get it. I'll definitely be looking into these 'germination indices', seems like it would be time well spent.

Many thanks to you.