RFP Tab 2 · Data Analysis · Tab 3(d) equity strategy and methodology

Two independent tests of every salary, with the Committee setting the rules.

The first test rebuilds what each member would be paid if CSN's own placement rules were applied consistently. The second fits a statistical model to the whole faculty and asks which salaries the model cannot explain. This page runs both on a synthetic file of 486 faculty so you can move the Committee's settings and watch the results change.

Lens 1 · rule-based expected salary

What the agreement's rules would pay this person

For each member: the placement-table cell for their education column and credited years at hire, in current dollars, carried forward by a credit for each year of CSN service. The size of that service credit is a Committee decision, because the table stops at ten years and the agreement does not say how service after hire should count. A member whose base salary sits more than a set tolerance below the expected salary is flagged.

Lens 2 · statistical model

What the whole faculty's pay pattern predicts

An ordinary least squares regression of log base salary on education column, credited years at hire, and CSN service (with a curve term, because pay growth slows over a career). The model is fitted to every member at once. A member whose actual salary falls well below the prediction (standardized residual below -1.5) is flagged. Agreement between the two lenses is the strongest evidence; disagreement is a question for the Committee.

Synthetic data · 486 generated records · no CSN employee

Equity demonstrator

The file was generated with four known problems planted in it: members placed before the 2022 placement formula, members whose column advance was only partly reflected in pay, members whose prior experience exceeded the ten-year cap, and a few market or difficult-to-recruit hires paid above the rule. Move the settings to see how each Committee choice changes who is flagged and what it costs.
0
Members flagged by the rule
0
Also flagged by the model
$0
Cost to close every flagged gap, with benefits
0
Cases fully closed with the year-one funding
0
Model R-squared on the synthetic file

Benefits and funding settings here are placeholders. In the study, the benefits rate comes from CSN's actual employer contributions and the funding figure from the amount CSN identifies by June 15 (CBA Art. 7 Sec. 4).

What the Committee decides

  • How a year of CSN service counts after hire
  • How much shortfall is tolerated before an adjustment
  • Whether adjustments go to the expected salary or to the tolerance floor
  • How market hires (Rank O) and difficult-to-recruit placements are treated
  • The order of funding when money is short

What the data must carry

  • Hire date and contract type (A, B, B+)
  • Column and row at hire, and credited years
  • Highest degree and date; every column move since hire
  • Current base salary, net of B+ and stipends
  • Tenure date, merit history and any prior equity adjustment

What we check before anyone sees a result

  • Every record ties to the payroll total HR provides
  • Credited years never exceed ten on the table
  • B+ salaries are normalized to base before comparison
  • Model diagnostics: residual spread by column and by service band
  • Each flagged case reviewed by a second analyst

An optional third check the Committee may want

The agreement asks about experience and education. Because CSN is a public employer, we also offer the Committee a test for pay differences by sex and by race or ethnicity that experience and education do not explain, reported to the Committee only. It uses the same model with group indicators added. If the Committee does not want it, it is not run.