Basic formulas of SURE model evaluation

Goal structure
$$C=\bigcap_{i=1}^rB_i=\bigcap_{i=1}^r \bigcup_{j=1}^{s_i} A_{ij} $$
$B_i$ - key goal, $i=1,...,r$
$A_{ij}$ - sub goal, $i=1,...,r$, $j=1,...,s_i$

Theoretical scores
$Q(A_{ij})$ - theoretical score of $A_{ij}$ $\Rightarrow$

Score of $C$
$$Q(C)=\prod_{i=1}^rQ(B_i) =\prod_{i=1}^r\big(1-\prod_{j=1}^{s_i}(1-Q(A_{ij})\big) $$
Evaluation score of $C$
$$Q_e(C)=\sqrt[\Large r]{\prod_{i=1}^rQ_e(B_i)} =\sqrt[\Large r]{\prod_{i=1}^r\bigg(1- \sqrt[\Large s_i]{\prod_{j=1}^{s_i}(1-Q(A_{ij})}\bigg)} $$

Empirical scores

$[x_0,x_1]$ - evaluation interval, $x_0 < x_1$.
Consider a sample of size $n$ of checklist results:
$$\quad x_{11}^{(1)},...,x_{1s_1}^{(1)},...,x_{r1}^{(1)},...,x_{rs_r}^{(1)}$$ $$...$$ $$\quad x_{11}^{(n)},...,x_{1s_1}^{(n)},...,x_{r1}^{(n)},...,x_{rs_r}^{(n)}$$
$x_{ij}^{(k)}$ - observed checklist score value for $A_{ij}$ of $k$-th checklist data record, $k=1,...,n$

Empirical scores for $k$-th checklist data record, $k=1,...,n$:
$$ Q^{*(k)}(A_{ij}) = q_{ij}^{*(k)} = \frac{x_{ij}^{(k)}-x_0}{x_1-x_0} $$
$$ Q^{*(k)}(B_i) = 1-\prod_{j=1}^{s_i}\big(1-q_{ij}^{*(k)}\big) $$ $$ Q^{*(k)}(C) = \prod_{i=1}^r\Big(1-\prod_{j=1}^{s_i}\big(1-q_{ij}^{*(k)}\big)\Big) $$ Empirical scores for total sample: Arithmetic mean over empirical scores for checklist data records for $k=1,...,n$. $$ Q^*(A_{ij})= \frac{1}{n}\sum_{k=1}^n Q^{*(k)}(A_{ij}) $$ $$ Q^*(B_i)= \frac{1}{n}\sum_{k=1}^n Q^{*(k)}(B_i) $$
$$ Q^*(C) = \frac{1}{n}\sum_{k=1}^n Q^{*(k)}(C) = \frac{1}{n}\sum_{k=1}^n\prod_{i=1}^r \Big(1-\prod_{j=1}^{s_i}\big(1-q_{ij}^{*(k)}\big)\Big) $$


Empirical evaluation scores

Empirical evaluation scores (empirical calibrated scores) for $k$-th checklist data record, $k=1,...,n$: $$ Q_e^{*(k)}(A_{ij})= Q^{*(k)}(A_{ij}) = q_{ij}^{*(k)} $$ $$ Q_e^{*(k)}(B_i)=1-\sqrt[\Large{s_i}]{ \prod_{j=1}^{s_i}\big(1-q_{ij}^{*(k)}\big)} $$ $$ Q_e^{*(k)}(C)= \sqrt[\Large r]{ \prod_{i=1}^r\bigg( 1-\sqrt[\Large s_i]{\prod_{j=1}^{s_i}\Big(1-q_{ij}^{*(k)}\Big)} \bigg) } $$

Empirical evaluation scores for total sample: Arithmetic mean over empirical evaluation scores of checklist data records for $k=1,...,n$. $$ Q_e^*(A_{ij}) =\frac{1}{n}\sum_{k=1}^n Q_e^{*(k)}(A_{ij}) =\frac{1}{n}\sum_{k=1}^n q_{ij}^{*(k)} =Q^*(A_{ij}) $$ $$ Q_e^*(B_i) =\frac{1}{n}\sum_{k=1}^n Q_e^{*(k)}(B_i) $$
$$ Q_e^*(C)= \frac{1}{n}\sum_{k=1}^nQ_e^{*(k)}(C) =\frac{1}{n}\sum_{k=1}^n \sqrt[\Large{r}]{ \prod_{i=1}^r\bigg( 1-\sqrt[\Large{s_i}]{\prod_{j=1}^{s_i}\Big(1-q_{ij}^{*(k)}\Big)} \bigg) } $$


Some properties of empirical evaluation score:
$$ q_{11}^{(k)}=\cdots =q_{rs_r}^{(k)}=1 \quad\Rightarrow\quad Q_e^{*(k)}(C)=1 $$ $$ Q_e^{*(k)}(B_i)=0\quad\Rightarrow\quad Q_e^{*(k)}(C)=0 $$ $$ q_{11}^{(k)}=\cdots =q_{rs_r}^{(k)}=q^* \quad\Rightarrow\quad Q_e^*(C)=q^* $$ $\quad\Rightarrow\quad$ The empirical evaluation score $Q_e^*(C)$ describes an 'average' checklist score level.


190328