turfR (version 0.8-7)

turf: TURF Analysis for R

Description

Calculate weighted reach and frequency statistics from TURF data for all possible combinations of n choose k, for user-specified combination sets, or for Monte Carlo simulated subsets of combinations.

Usage

turf(data, n, k, combos, ...)

Arguments

data

Required. Literal character string representing name of a file in the working directory readable using read.table(data, header=TRUE), or name of a data frame or matrix in R containing TURF data. Rows are individuals (respondents). Col

n

Required. Scalar indicating the number of items to be included in the TURF algorithm; 0 < n < (ncol(data) - 1). Non-integer values are coerced using floor(n).

k

Required. Vector of length 1 to n containing any values 1 to n indicating the combination sizes to be evaluated by the TURF algorithm. Non-integer values are coerced using floor(k).

combos

Optional. List of combination sets to be evaluated by the TURF algorithm, such as that generated by turf.combos. Individual combination sets are p x n matrices, containing only zeroes and ones indicating items to be in

...

Optional. Additional arguments controlling behavior and output of the TURF algorithm. See turf.args. Arguments indicated here must match named arguments in turf.args.

Value

• R object: A list of 2 elements
• turf
  A list of length(k) elements, each of which is comprised of a matrix whose columns are reach (rchX), frequency (frqX), and n indicator variables representing the items included in a given combination. Rows represent combinations evaluated. Sorting and the number of rows returned are controlled by the sort and keep arguments passed from turf.args.
• call
  The call to turf as a literal character string
• Written to Console: Combination size(s) evaluated, as k of n; time in seconds required to evaluate all combinations of a given size; total time in seconds required to complete the function call.

Details

TURF algorithm is as originally described by Miaoulis, et al. (1990) and outputs reach and frequency statistics for each combination evaluated. Reach = sum(weights[reached]) / sum(weights), where an individual is considered "reached" if the reach criterion is met for at least the number of items indicated by the depth argument in turf.args. See Markowitz (2005) for a more detailed explanation of depth of reach. Frequency = sum(weights x items_reached) / sum(weights). Frequency includes all individuals, whether reached or not. Frequency among "reached" individuals may be calculated as Frequency / Reach regardless of weights. When the combos argument is omitted from the call, combination sets are generated automatically using turf.combos and information passed to it by n, k and any additional arguments passed from turf.args. The user may also pass any combination sets (e.g., a user-truncated set) developed in or out of R, so long as the structure of the combination sets corresponds to that described above. Monte Carlo simulated combination sets, when requested, are generated using the procedure described by Adler, et al. (2010). Sets of n choose k that result in small numbers of combinations should be run without the Monte Carlo simulation option since they require little processing time (e.g., 12 choose 7 = 792 combinations and will run in less than 1 second). Typically, even larger sets will run in a reasonable amount of time without subsetting (e.g., 18 choose 10 = 43,758 combinations and will run in less than a minute on most computers), but RAM may become a limiting factor, especially when large numbers of individuals are combined with problems involving large numbers of items. The default for substituting Monte Carlo simulated subsets of combinations is set at 10,000. For the above reasons, and because subsetting can lead to duplicate combinations especially when the original set is small, care should be taken to not subset combinations until the original set becomes unmanageably large. The lack of looping in the algorithm permits processing at about 1,000 combinations per second.

References

Adler, T.J., Smith, C. & Dumont, J. 2010. Optimizing product portfolios using discrete choice modeling and TURF. In: S. Hess, A. Daly (Eds), Choice modeling: the state-of-the-art and the state-of-practice; proceedings from the Inaugural International Choice Modeling Conference. Emerald Publishing Group Ltd., pp. 485-497. Krieger, A.M. & Green, P.E. 2000. Turf revisited: Enhancements to toal unduplicated reach and frequency analysis. Marketing Research, 12, 30-36. Markowitz, L. 2005. Going beyond TURF to complement and extend existing product lines. Ipsos-Insight, November 2005. Miaoulis, G., Free, V. & Parsons, H. 1990. TURF: A new planning approach for product line extensions. Marketing Research, March, pp. 28-40.

Examples

##Example 1
  ##Evaluate all combinations of 3, 4, 5 and 6 items drawn from 10 items.
  data(turf_ex_data)
  ex1 <- turf(turf_ex_data, 10, 3:6)
  
  ##Output to Console:
  ##3 of 10: 0.105068 sec
  ##4 of 10: 0.1420949 sec
  ##5 of 10: 0.1511021 sec
  ##6 of 10: 0.1160791 sec
  ##total time elapsed: 0.518347 sec
  
  
  ##Example 2
  ##Pass additional arguments 
  data(turf_ex_data)
  ex2 <- turf(turf_ex_data, 10, 3:6, depth=2, keep=20, mc=TRUE, nsims=100) 
  
  ##Output to Console:
  ##3 of 10: 0.03802586 sec
  ##4 of 10: 0.03702521 sec
  ##5 of 10: 0.0380249 sec
  ##6 of 10: 0.03802609 sec
  ##total time elapsed: 0.156105 sec 
  
  
  ##Example 3
  ##Customize combos, exclude item 10 from all combinations of size 3
  data(turf_ex_data)
  psims <- colSums(turf_ex_data[,-c(1:2)]) / sum(turf_ex_data[,2])
  psims <- psims / sum(psims)
  combos <- turf.combos(10, 3, mc=TRUE, nsims=100, psims=psims)
  combos[[1]] <- combos[[1]][-which(combos[[1]][,10]==1),]
  ex3 <- turf(turf_ex_data, 10, 3, combos)
  
  ##Output to Console:
  ##3 of 10: 0.02001405 sec
  ##total time elapsed: 0.02001405 sec