I try to calculate the split-half reliability for my item-set. My data look like this (names are slightly different, actually it is not "Item2" but "2" etc.):
Participant Item2 Item3 Item4 etc. 1 0 1 1 2 1 0 NA etc.
0 means that the person resolved the sentence in a neutral and 1 in an emotional way. NA means that the participant did not find a correct solution.
With the following code I was trying to calculate the split-half reliability:
nBoot <- 10000 # defining the resamples
output <-psych::splitHalf(dat_wide[c("2" , "3" , "4" , "5" , "7" , "8" , "9" ,"10", "11" ,"12" ,"13", "14" ,"15" ,"16" ,"17" ,"18" ,"19", "20" ,"21" ,"22", "24" ,"27" ,"28", "30" ,"31" ,"35" ,"36", "37" ,"39" ,"40" ,"41" ,"42" ,"44" ,"47" ,"49" ,"50", "52" ,"53" ,"54", "55" ,"56", "57" ,"58" ,"59", "60", "61", "62" ,"63" ,"64" ,"65" ,"66", "67", "68" ,"69" ,"70", "71", "72" ,"73", "74", "75")],
raw = T,
n.sample = nBoot
# ,check.keys = FALSE
)
If I dont switch off check.keys I get the following error messages:
1: In cor.smooth(r) : Matrix was not positive definite, smoothing was done
2: Ist das ein Problem: In psych::splitHalf(dat_wide[c("2", "3", "4", "5", "7", "8", "9", :Some items were negatively correlated with total scale and were automatically reversed.
By reading answers and comments on the first error message I understand that I probably dont have to worry about it since it is a result of the 0/1 coding.
However the second error message worries me, since I do not want to include negatively correlated items in my item-set. I heard that one can see somewhere in the output which items are those negatively correlated items. However, I can not find it in the output.I tried "output$raw", but as far as I understand these are the correlations of first and second test half of all the resamples and not of the items.
I would be extremly grateful if someone could tell me how to figure out which Items are negatively correlated so I can figure out if I want to remove them from my Itemset.
Thank you so much in advance!
As it might be helpful for answering I also include the complete output I get:
In smc, smcs > 1 were set to 1.0
Warning messages:
1: In cor.smooth(r) : Matrix was not positive definite, smoothing was done
2: In psych::splitHalf(dat_wide[c("2", "3", "4", "5", "7", "8", "9", :
Some items were negatively correlated with total scale and were automatically reversed.
> output
Split half reliabilities
Call: psych::splitHalf(r = dat_wide[c("2", "3", "4", "5", "7", "8",
"9", "10", "11", "12", "13", "14", "15", "16", "17", "18",
"19", "20", "21", "22", "24", "27", "28", "30", "31", "35",
"36", "37", "39", "40", "41", "42", "44", "47", "49", "50",
"52", "53", "54", "55", "56", "57", "58", "59", "60", "61",
"62", "63", "64", "65", "66", "67", "68", "69", "70", "71",
"72", "73", "74", "75")], raw = T, n.sample = nBoot)
Maximum split half reliability (lambda 4) = 0.99
Guttman lambda 6 = 1
Average split half reliability = 0.91
Guttman lambda 3 (alpha) = 0.91
Guttman lambda 2 = 0.92
Minimum split half reliability (beta) = 0.77
Average interitem r = 0.15 with median = 0.15
2.5% 50% 97.5%
Quantiles of split half reliability = 0.85 0.91 0.96