PCA analyse

[aska_ujita]

Hello there,

I am testing cattle behavior in two groups... One got habituation protocol and one is the control (no habituation).

I collected behavioral parameters: 

Variable	Description	Scores
1	2	3	4	5
STOP	If the animal halted at the entrance of the chute	No	Yes, calmly	Yes, in fear	-	-
HELP	If the animal received help to enter the chute	No	Voice command	Pat	One poke with an instrument*	Two or more pokes with an instrument*
ENTS	Speed to enter the chute	Walked	Trotted	Ran	-	-
VOC	If the animal vocalized	No	Communication	Fear	Pain	-
RUM	If the animal ruminated	No	Yes	-	-	-
SNORT	If the animal snorted	No	Yes	-	-	-
DISP	If the animal displaced its legs during brushing	No, it was calm	Yes, medium displacement	Yes, very unquiet	-	-
MOV	If the animal moved its body during brushing	No, it was calm	Yes, medium movement	Yes, very unquiet	-	-
EAR	If the animal moved its ears during brushing	No, it was calm	Yes, medium movement	Yes, very unquiet	-	-
EXT	Speed to exit the chute	Walked	Trotted	Ran	-	-

I would like to make a figure similar to this one below for all these parameters above:

Can someone help me in this? 

Thank you!

Kind regards, Aska.

[PaigeMiller]

You can't do PCA on data that is categorical or text.

I would like to make a figure similar to this one below for all these parameters above:

Given that you can't do PCA, you can't get such a graph, which comes from PCA.

So let's take a step back. Please describe what you would like to learn from your data, then we can figure out what analysis to perform and what graphics or tables might help.

[aska_ujita]

Hi PaigeMiller, 

Yes, I understand that. I thought maybe to use the frequency of it (I have it from PROC FREQ).

Do you think is possible??

I would like to understand if the studied behaviors and classes used are coherent.

Thank you!!

[PaigeMiller]

---

@aska_ujita wrote:

Hi PaigeMiller, 

Yes, I understand that. I thought maybe to use the frequency of it (I have it from PROC FREQ).

Do you think is possible??

---

I don't think its a good idea to decide what statistics to use until its clear what you want to learn from the data.

I would like to understand if the studied behaviors and classes used are coherent.

And perhaps because I am not an animal behavior scientist, I don't know what this means. Can you give a more detailed and concrete example?

[ballardw]

Since your graph image shows categories that do not apparently exist in your data then how to get to those categories might one issue. Or is the "text" shown just an example and does not match your data at all?

[aska_ujita]

Hi ballardw, this is just an example.

I just used the graphic as an example, I would like to make a similar but with my behavioral parameter (in Table).

[aska_ujita]

Hi PaigeMiller, yes, I understand what you mean... but my advisor asked me that.

I never did this before and I even dont understand for what is this exactly... But I am saying exactly what he told me.

Maybe I can try just with the frequency of these data?

[PaigeMiller]

I would ask your advisor some additional questions. 

The good thing about SAS is also that you can perform just about any analysis on any data; but this is also the bad thing about SAS, if you don't put any thought into it or don't understand what the problem requires, SAS gives you totally meaningless results.

[aska_ujita]

True story.

Yes, I will ask him, thank you!

[Rick_SAS]

I'm going to guess that you are thinking about PROC PRINQUAL, which is like PCA for ordinal data. It can make a plot for nominal data that looks similar to what you displayed. You can read about the difference between the assumptions in each procedure. PROC PRINQUAL can be used to create biplots, but I think you can also suppress the scatter plot and just look at the vector plot of variables. 

You can read more about biplots and how to create one by using PROC PRINQUAL