hrsRcePred {dbd}

Horse race prediction data.

Description

Counts of correct predictions of the outcomes of 10 harness races made by “experts” and “non-experts”.

Usage

hrsRcePred

Format

A data frame with 30 observations on the following 4 variables.

sbjType

A character vector with entries "NonXpert" and "Expert", which classifies the “subjects” (the people making the predictions of the race outcomes).

subject

An integer vector indexing the subjects. (Not of any real consequence.)

top1

An integer vector giving the counts of correct predictions of the winners of 10 harness races.

top3

An integer vector giving the counts of correct predictions of the top three horses (“win/place/show” in 10 harness races.

Details

In Ceci and Liker (1986) it is stated that subjects were classified as “experts” and “nonexperts” based on their ability to predict post-time odds on the basis of factual information about horses.

It appears that the counts in top1 and top3 pertain to the same 10 races, but this is not completely clear.

Source

These data are taken from the paper cited in the first of the two given in the References below. They were provided by a generous email correspondent who prefers to remain anonymous.

References

Ceci, S. J. and Liker, J. K. (1986). A day at the races: A study of IQ, expertise, and cognitive complexity. Journal of Experimental Psychology, General 115, pp. 255 – 266.

Ceci, S. J. and Liker, J. K. (1988). Stalking the IQ-expertise relation: When the critics go fishing. Journal of Experimental Psychology, General 117, pp. 96 – 100.

Examples

    X     <- hrsRcePred
    top1e <- X[X$sbjType=="Expert","top1"]
    top1n <- X[X$sbjType=="NonXpert","top1"]
    top3e <- X[X$sbjType=="Expert","top3"]
    top3n <- X[X$sbjType=="NonXpert","top3"]
    dbfit1e <- mleDb(top1e,ntop=10,zeta=TRUE)
    dbfit1n <- mleDb(top1n,ntop=10,zeta=TRUE)
    dbfit3e <- mleDb(top3e,ntop=10,zeta=TRUE)
    dbfit3n <- mleDb(top3n,ntop=10,zeta=TRUE)
# Set seeds to get repeatable Monte Carlo p-values.
    ## Not run:  # Takes too long.
        print(gof(dbfit1e,obsd=top1e,MC=TRUE,maxit=5000,verb=TRUE,seed=49)$pval) # 0.02
        print(gof(dbfit1n,obsd=top1n,MC=TRUE,verb=TRUE,seed=128)$pval)           # 0.79
        print(gof(dbfit3e,obsd=top3e,MC=TRUE,verb=TRUE,seed=303)$pval)           # 0.35
        print(gof(dbfit3n,obsd=top3n,MC=TRUE,maxit=3000,verb=TRUE,seed=24)$pval) # 0.40
    
## End(Not run)
    bbfit1e <- mleBb(top1e,size=10)
    bbfit1n <- mleBb(top1n,size=10)
    bbfit3e <- mleBb(top3e,size=10)
    bbfit3n <- mleBb(top3n,size=10)
# Set seeds to get repeatable Monte Carlo p-values.
    ## Not run:  # Takes too long.
        print(gof(bbfit1e,obsd=top1e,MC=TRUE,verb=TRUE,seed=792)$pval) # 0.11
        print(gof(bbfit1n,obsd=top1n,MC=TRUE,verb=TRUE,seed=48)$pval)  # 0.64
        print(gof(bbfit3e,obsd=top3e,MC=TRUE,verb=TRUE,seed=969)$pval) # 0.62
        print(gof(bbfit3n,obsd=top3n,MC=TRUE,verb=TRUE,seed=834)$pval) # 0.75
    
## End(Not run)
# Reality check: goodness of fit tests for the fit of just plain *binomial*
# distributions to these data sets yielded Monte Carlo p-values equal to
# 0.22, 0.17, 0.32 and 0.73 respectively.  I.e. binomial fits appear to
# work just fine!