Update Ch3.ipynb

Ch3.ipynb

@@ -166,23 +166,6 @@
     "fit!(mm₁)"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "We can answer the question of whether Costello and Watts' proposal can account for Douven and Schupbach's findings by trying to predict the subjective updated probabilities reported by Douven and Schupbach not by the objective conditional probabilities but instead by the 'noisy' version of that predictor, transformed according to the formula given by Costello and Watts. The following first defines a function that takes Costello and Watts' error parameter *d* as input and outputs the transformed predictor for the given value of *d*. The function is based on Eq. 17 in Costello and Watts (2018), according to which\n",
-    "$$\n",
-    "\\Pr\\nolimits_*(A\\,|\\,B) \\:\\: = \\:\\: \\frac{(1-2d)^2\\Pr(A\\wedge B) + d(1-2d)\\bigl(\\Pr(A) + \\Pr(B)\\bigr) + d^2}{(1-2d)\\Pr(B) + d},\n",
-    "$$\n",
-    "where $\\Pr_*(A\\,|\\,B)$ is the noisy estimate of the probability of $A$ conditional on $B$, and with the noise parameter $d\\in[0,.5)$. Notice that if $d=0$, indicating that there is no noise, then \n",
-    "$$\n",
-    "\\Pr\\nolimits_*(A\\,|\\,B)  \\:\\: = \\:\\: \\frac{\\Pr(A\\wedge B)}{\\Pr(B)} \\:\\: = \\:\\: \\Pr(A\\,|\\,B).\n",
-    "$$\n",
-    "Then we fit for *d* going from 0 to .5, in small increments, mixed models like the ones above except that the new models have the noisy objective probabilities, rather than the unnoisy ones, as predictor. We look whether for at least one value of *d* the correspondig Bayesian model does best, or whether explanatory considerations do still make a significant contribution to model fit even when noise is taken into account.\n",
-    "\n",
-    "First calculate the probabilities needed to perform the Costello and Watts transformation:"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -217,13 +200,6 @@
     "first(data1, 6)"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "The following function calculates AIC values for the Bayesian model (with 'noisified' objective conditional probabilities as predictor), the model with noisified objective conditional probabilities and judgments of explanatory goodness as predictors, and the model with noisified objective conditional probabilities and *difference* in explanatory goodness as predictors, for a range of possible noisifications:"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,