Analysing difference between 20+ variables when each only have one value.

I'm currently analysing data I have collected looking at the rate of interaction between a species of shark and 24 different teleost fish species.

My variables are:
x = species and y = rate of aversion

In trying to calculate their statistical difference from one another I have become confused as to which test to use.

When trying to conduct an Anova or Kruskal-Wallis test (due to the data's non-normality) followed by a Tukey HSD test, I ran up against the problem that they compare means and, as i only have one measurement per group, I am met with 'NaNs' as my output.

If someone could guide me to or advise me on the correct test for this data, I would be evry thankful.

My data is arranged like this:
species ----------aversion_rate (per individual)------group size
|Orangespot_surgeonfish ----------- 0.5-----------------------2
|Yellowback_fussilier ----------------6.09----------------------58
|Threadfin_butterflyfish ------------3.5------------------------2
|Kleins_Butterflyfish --------------- 1.5------------------------0
|Racoon_butterflyfish --------------0.49-----------------------6
|Copperband_butterflyfish --------4.5-------------------------2
|Striped_large-eye_bream ---------1.21-----------------------9
|French_grunt -------------------------0.4------------------------5
|Barred_flagtail -----------------------0.75----------------------21
|Cleaner_wrasse ----------------------0-------------------------2
|Humpback_snapper ---------------3.25----------------------4
|Bluelined_snapper ----------------0.39-----------------------15
|Stripey --------------------------------1.47-----------------------21
|Blackbar_soldierfish ---------------3.8------------------------5
|Unicorn_tang -----------------------1----------------------------2
|Bignose_unicornfish ---------------0--------------------------2
|Red-toothed_triggerfish ----------3--------------------------3
|Harlequin_sweetlips ---------------2--------------------------4
|Yellowbanded_sweetlips-----------5--------------------------2
|Emperor_angelfish ------------------2--------------------------1
|Yellowface_angelfish ----------------3--------------------------1
|Reef_squirrelfish ----------------------9.5-----------------------4
|Sabre_squirrelfish --------------------0.8------------------------5
|Foxface_rabbitfish -------------------0.91-----------------------13

If you dont have data relating to the volume of measurements taken, I dont see how its possible for you to calculate meaningful further statistics.

Perhaps you have them in aother file(s) though ?

Apologies, the aversion rate was counted for each species and i subsequently divided this count by the number of individuals of each group so as to standardise my data. I have since included the group size for each recorded species.

because aov wraps lm it supports a weights argument, so I would think you could do aov, and then the Tukey

I initially thought the same but after researching I became aware that as i only have one sample per group (species), the anova wouldn't run as it tests for a variance in means.

I then believed this is why my TukeyHSD outcome looks like this:

Fit: aov(formula = aversion_rate ~ species)
$species
--------------------------------------------------------------- diff lwr upr
Bignose_unicornfish-Barred_flagtail ----- -7.500000e-01 NaN NaN
Blackbar_soldierfish-Barred_flagtail ------ 3.050000e+00 NaN NaN
Bluelined_snapper-Barred_flagtail ------- -3.600000e-01 NaN NaN
Cleaner_wrasse-Barred_flagtail ----------- -7.500000e-01 NaN NaN
Copperband_butterflyfish-Barred_flagtail ---- 3.750000e+00 NaN NaN

etc.

I may be wrong in this conclusion.

I additionally thought that, even if it did work there may be a better test as having 576 p-values wouldn't really tell me anything about my data?

This topic was automatically closed 42 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.