Dear all,

I am using a Logit Regression to see if companies did recover to their pre-crisis ROA levels (yes if recovery within certain window, no if not). To check the assumptions of linearity i used a Box-Tidwell Test which showed me by its p-values that the ROA_2007Q3 as well as the crisis_drop_size the companies saw in their ROA are non-linear below a 5% level. I relied on this homepage https://stats.idre.ucla.edu/stata/we...n-diagnostics/ to interpret my findings and in line with their example i used the squareroot of the drop_size and obtained even worse results. How do i need to change my variable definitions of ROA_2007Q3 and drop_3_qbq (the drop size) to make them linear with the DV?

With this code i get the below output (ROA and drop are right now represented as % or %points):

boxtid logit strict_recovery_dummy_3_qbq femaleCEO CEOtenure CEOduality CEOage PotentialSlackResources PercentageFemaleDirectors ROA_2007Q3 Q1_dummy Q2_dummy Q3_dummy AgricultureForestryFishing Mining Construction TransportationPublicTransport WholesaleTrade RetailTrade FinanceInsuranceRealEstate Services Nonclassifiable drop_size

[Total iterations: 60]

Box-Tidwell regression model

note: IROA__p1 omitted because of collinearity.

Logistic regression Number of obs = 933
LR chi2(25) = 151.61
Prob > chi2 = 0.0000
Log likelihood = -489.04352 Pseudo R2 = 0.1342


strict_recovery_dumm~q Coefficient Std. err. z P>z [95% conf. interval]

ICEOt__1 -.2355592 2.286112 -0.10 0.918 -4.716255 4.245137
ICEOt_p1 -.0018722 1.071501 -0.00 0.999 -2.101976 2.098232
ICEOa__1 -22.71017 124.3354 -0.18 0.855 -266.403 220.9827
ICEOa_p1 .0121845 106.9557 0.00 1.000 -209.6171 209.6415
IPote__1 1.684909 2.507962 0.67 0.502 -3.230607 6.600424
IPote_p1 -.0205716 .824244 -0.02 0.980 -1.63606 1.594917
IPerc__1 -.0792827 .1630877 -0.49 0.627 -.3989287 .2403633
IPerc_p1 .0016973 .1076719 0.02 0.987 -.2093357 .2127304
IROA___1 3.697561 .7425445 4.98 0.000 2.242201 5.152921
IROA__p1 0 (omitted)
Idrop__1 .3659422 .3068437 1.19 0.233 -.2354604 .9673448
Idrop_p1 .0025486 .0656988 0.04 0.969 -.1262187 .1313158
femaleCEO .0706754 .455044 0.16 0.877 -.8211945 .9625453
CEOduality -.191278 .1716724 -1.11 0.265 -.5277496 .1451937
Q1_dummy -.432766 .2018574 -2.14 0.032 -.8283992 -.0371328
Q2_dummy -.649213 .2344642 -2.77 0.006 -1.108754 -.1896717
Q3_dummy -.1413489 .2522684 -0.56 0.575 -.6357859 .3530882
Agriculture -1.212988 1.446473 -0.84 0.402 -4.048023 1.622047
Mining -1.408233 .359642 -3.92 0.000 -2.113119 -.7033478
Construction -.3813196 .7636673 -0.50 0.618 -1.87808 1.115441
Transpo -1.220585 .278687 -4.38 0.000 -1.766801 -.6743682
Wholesale -.7396785 .4532187 -1.63 0.103 -1.627971 .1486138
RetailTrade .9028717 .4128874 2.19 0.029 .0936272 1.712116
Finance -.5037107 .2687348 -1.87 0.061 -1.030421 .0229999
Services -.0940663 .2456208 -0.38 0.702 -.5754743 .3873417
Nonclass -.8409565 1.256167 -0.67 0.503 -3.302998 1.621085
_cons 1.242205 .2360469 5.26 0.000 .7795617 1.704848

Note: 0 failures and 4 successes completely determined.

CEOtenure -.1260785 .1280936 -0.98 Nonlin. dev. 0.007 (P = 0.932)
p1 .6569044 4.304968
CEOage .0233692 .0126516 1.85 Nonlin. dev. 0.570 (P = 0.450)
p1 -2.141594 4.70045
PotentialS~s .0439721 .0412345 1.07 Nonlin. dev. 3.313 (P = 0.069)
p1 .2163807 .49866
Percentage~s -.0094062 .0058035 -1.62 Nonlin. dev. 0.015 (P = 0.901)
p1 1.139591 1.591438
ROA_2007Q3 -.1875269 .0392912 -4.77 Nonlin. dev. 13.566 (P = 0.000)
p1 -1 0
drop_size -.0080668 .0089612 -0.90 Nonlin. dev. 44.008 (P = 0.000)
p1 -.4698967 .1961628

Deviance: 978.089.


here is also the Picture of the output attached for nicer readability

Array