Cognitive Models logo Cognitive Models

The method ggdmc adapts different factorial designs is to use Boolean model matrices, which associate experimental conditions with latent variables / model (free) parameters. The model parameters are often designed to account for cognitive operations that cannot be directly observed. Three examples are the rate of the degradation of memory strength, the rate of (sensory) evidence accumulation, and the response threshold.

Take regression models for example. One might be interested in examining intercepts and slopes, the two regression model parameters by themselves usually do not carry psychological meanings. Of course, one can construct a framework to harness the (regression) model parameters. For example, in traditional visual search studies, mean response times (MRTs) are often associated with the display sizes and the slopes of the MRT-Display size function were conveniently interpreted as search efficiency (Treisman & Gelade, 1980). This was useful strategy as a staring point, but needs further refinement to get more insights (e.g., to understand speed-accuracy trade-off issue, serial vs. parallel processing etc.).

It is therefore and often needed to refine the basic regression model to further accommodate many intricate cognitive constructs. ggdmc hard-wires, the diffusion decision and the linear ballistic models and applies the method of Boolean matrices to serve the purpose of adapting factorial designs and that of accounting for latent variables of RT models.

The first step in ggdmc is to set up a 3-D model array.

Build Models

BuildModel creates a model array, which composes of many model matrices. Each of them represents a response. The content of a model matrix indicates the correspondence of parameters and design cells. For example, if a data set has a two-level stimulus factor, affecting the drift rate (as in DDM), a model matrix will have two drift rate parameters, say, v.d1 and v.d2 (d stands for difficulty). One could understand this idea of correspondence between an experimental factor and its parameter mapping by examining the following example.

Example 1

In this example, I used the LBA model (Brown and Heathcote, 2008) to illustrate, fitting data from a single participant. The LBA’s B parameter depends only on response (R). The mean and the standard deviation of the drift rates depends on M (matching) factor. The experimental design has one two-level stimulus factor (S). The following model presumes the S factor has no effect on any model parameter. The accuracy is determined by S and R. The M factor is a specific latent factor just for the LBA model.

 model <- BuildModel(
   p.map     = list(A = "1", B = "R", t0 = "1", mean_v = "M",
                    sd_v = "M", st0 = "1"),
   match.map = list(M = list(s1 = "r1", s2 = "r2")),
   factors   = list(S = c("s1", "s2")),
   constants = c(sd_v.false = 1, st0 = 0),
   responses = c("r1", "r2"),
   type      = "norm")

p.map means parameter map. match.map matches the stimulus type to the response type to determine if a response is correct or error. factors means experimental factors, constants specifies which model parameter to fix as constant values. This is to enforce model assumptions. responses indicates response types, by specifying character strings or numbers. Lastly, type specifies the model types, such as the diffusion decision model (rd) or the LBA (norm).

For illustration purpose, I simulated some realistic response time data. I made up a true parameter vector. This is usually unknown and estimated from data.

p.vector  <- c(A = .75, B.r1 = .25, B.r2 = .15, t0 = .2, mean_v.true = 2.5,
               mean_v.false = 1.5, sd_v.true = 0.5)

print will show the model array together with its attributes that have been added into in the BuildModel step.

print(model)
## r1 
##          A B.r1  B.r2   t0 mean_v.true mean_v.false sd_v.true sd_v.false  st0
## s1.r1 TRUE TRUE FALSE TRUE        TRUE        FALSE      TRUE      FALSE TRUE
## s2.r1 TRUE TRUE FALSE TRUE       FALSE         TRUE     FALSE       TRUE TRUE
## s1.r2 TRUE TRUE FALSE TRUE        TRUE        FALSE      TRUE      FALSE TRUE
## s2.r2 TRUE TRUE FALSE TRUE       FALSE         TRUE     FALSE       TRUE TRUE
## r2 
##          A  B.r1 B.r2   t0 mean_v.true mean_v.false sd_v.true sd_v.false  st0
## s1.r1 TRUE FALSE TRUE TRUE       FALSE         TRUE     FALSE       TRUE TRUE
## s2.r1 TRUE FALSE TRUE TRUE        TRUE        FALSE      TRUE      FALSE TRUE
## s1.r2 TRUE FALSE TRUE TRUE       FALSE         TRUE     FALSE       TRUE TRUE
## s2.r2 TRUE FALSE TRUE TRUE        TRUE        FALSE      TRUE      FALSE TRUE
## model has the following attributes: 
##  [1] "dim"        "dimnames"   "all.par"    "p.vector"   "pca"        "par.names" 
##  [7] "type"       "factors"    "responses"  "constants"  "posdrift"   "n1.order"  
## [13] "match.cell" "match.map"  "class"     

print, when supplied with a true parameter vector, will show how the factorial design is assigned to model parameters. Understanding the assigning process is an advanced topic. I will return to it at a different tutorial. Note that I, using Brown and Heathcote’s (2008) convention, differentiate the lowercase b and uppercase B in the LBA model. The former means the threshold parameter, and the latter is the travel distance parameter. The LBA model assumes b = A + B.

print(model, p.vector)
## "s1.r1"
##    A   b  t0 mean_v sd_v st0
## 0.75 1.0 0.2    2.5  0.5   0
## 0.75 0.9 0.2    1.5  1.0   0
## "s2.r1"
##    A   b  t0 mean_v sd_v st0
## 0.75 1.0 0.2    1.5  1.0   0
## 0.75 0.9 0.2    2.5  0.5   0
## "s1.r2"
##    A   b  t0 mean_v sd_v st0
## 0.75 0.9 0.2    1.5  1.0   0
## 0.75 1.0 0.2    2.5  0.5   0
## "s2.r2"
##    A   b  t0 mean_v sd_v st0
## 0.75 0.9 0.2    2.5  0.5   0
## 0.75 1.0 0.2    1.5  1.0   0