Nonlinear Interaction Models Nonlinear Interaction Models PS699, Winter 2010 (Here for further pedagogical slides & materials: http://www.umich.edu/~franzese/SyllabiEtc.html ) Robert J. Franzese, Jr. Professor of Political Science, The University of Michigan Ann Arbor The University of Michigan, Ann Arbor Slide 1 of 37
37
Embed
Nonlinear Interaction ModelsInteraction Modelsfranzese/ps699.NonlinearModelsAndNLS.pdf– Convince skeptic some causal effect exists vsConvince skeptic some causal effect exists, vs.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
– Ignore context conditionality (stay linear-additive):I ffi i b bi d ll d• Inefficient at best, biased more usually, and, anyway, context-conditionality is our interest!
– Isolate one or some very few interactions for close study; i ( li i i )ignore rest (stay linear-interactive):
• Same, to degree lessened by amount of interaction allowed, but demands on data rise rapidly w/ that amount.
– “Structured Case Analysis”:• May help ‘theory generation’, but, for empirical evaluation, doesn’t
help; worsens problem! (See Franzese OxfHndbk CP 2007).p; p ( f )— EMTITM: Lean harder on thry/subst to specify more precisely
the nature interax: functional form, precise measures, etc.• Refines question put to the data (changes default tests also)• Refines question put to the data (changes default tests also).• GIVEN thry/subst. specification into empirical model, can estimate
complex interactivity. Side benefits. But must give.Slide 7 of 37
Nonlinear Least-Squares & EMTI• EITM: Empirical Implications of Theoretical Models
– Vision: Theory ⇒ more, sharper predictions ⇒ better tests, which th f i f th hi htherefore inform theory more, which…
• TMEI: Theory-specified Models for Empirical Inference– Vision: Theory structures empirical models & relations b/w obs ⇒y p /
specification & (causal) i.d. of empirical models• TIEM: Theoretical Implications of Empirical Measures
– Vision: Emp regularities findings measures inform theory dev’pVision: Emp. regularities, findings, measures inform theory dev p.• EMTITM: Empirical Models of Theoretical Intuitions
– Vision: Intuitions derived from theoretical models specify empirical d l I i i l ifi i h i i i d lmodels. I.e., empirical specification to match intuitions, not model.
• Note: Strongly counter some alternative moves stats & econometrics, & related; there toward non-parametric, , ; p ,matching, & experimentation—there, “model-dependence” a 4-letter word. Alternative audiences & rhetorical purposes?– Convince skeptic some causal effect exists vsConvince skeptic some causal effect exists, vs.– For the convinced, give richer, portable model of how world works.
Slide 8 of 37
Complex Context-Conditionality and Nonlinear Least-SquaresNonlinear Least-Squares
• Complex Context-Conditionality: Effect of p yanything depends on most everything else. E.g.:– Policymaking:
• Socioeconomic-structure of interests• Party-system and internal party-structures
El t l t & G t l t• Electoral system & Governmental system• Socio-economic realities linking policies to outcomes
– Comparative Democratic Budgeteering e g beginsComparative Democratic Budgeteering, e.g., begins as this simple proposition…
( ) ( ) ( )B m s n= + × × +x x x– …and eventually becomes this…
... ( ) ( ) ( ) ...m s nB m s n+ × × +x x x
Slide 9 of 37
Complex Context-Conditionality and Nonlinear Least-SquaresNonlinear Least-Squares
• Comp Dem Budgeteering…eventually becomes this:
• …where incentive-nature given by...
... ( ) ( , ) ( , , , , , ) ...s rB s u m ec gc n d p u IPC DMρ= + × × +…where incentive nature given by...
ˆThat is, intuitively, writing ( , ) as simply , we hav
LS LS LS
LS
f fn kf
σ ′⎡ ⎤ ⎡ ⎤≡ = = − −⎣ ⎦ ⎣ ⎦−∇ ∇
2V( ) I V y X y X
Xβ
ε Ω β β
β1
e:ˆ ( )′ ′∇ ∇ ∇β
( )1
1 1 1
2
( )
ˆ ( ) ( ) ( ) ( ) ,ˆ ˆwhich if ( , ) meaning , &, if = , gives the famiar
LS
LS LS
S Sf σ
−
− − −
′ ′= ∇ ∇ ∇
⎡ ⎤′ ′ ′ ′ ′= ∇ ∇ ∇ = ∇ ∇ ∇ ∇ ∇ ∇⎣ ⎦= ∇ =
y
V V y V y
X X X I
β
β
β β Ω
• NLS is BLUE under same conditions as OLS, w/ ∇ for X.
1 2 1
which if ( , ) meaning , &, if , gives the famiar
ˆ ˆ( ) & ( ) ( ) , as always.
LS LS
LS LS LS
f σ
σ− −
∇
′ ′ ′= =
X X X I
X X X y V X X
β β Ω
β β
• Interpreting NLS (already know how): Effects = derivatives & 1st-differences; s.e.’s by Delta Method or simulation as usual…
Slide 11 of 37
Generalized Nonlinear Least-Squares2 2( ) i h ( )f VX Iβ Ω• GNLS:2 2
1 1 1
( , )+ with ( )ˆ ( )
GNLS
f V σ σ− − −
= = ≠′ ′⇒ = ∇ ∇ ∇
y X I
y
β ε ε Ωβ Ω Ω
1 1 1 1 1 1
1 1 1 1 1 1
ˆ( ) ( ) ( ) ( )
( ) ( )GNLS
V V− − − − − −
− − − − − −
′ ′ ′⇒ = ∇ ∇ ∇ ∇ ∇ ∇′ ′ ′= ∇ ∇ ∇ ∇ ∇ ∇
yβ Ω Ω Ω ΩΩ Ω ΩΩ Ω
GNLS is BLUE in same cond’s NLS but Ω for I
1 1 1 1 1 1 1
( ) ( )
( ) ( ) ( )− − − − − − −′ ′ ′ ′= ∇ ∇ ∇ ∇ ∇ ∇ = ∇ ∇Ω Ω Ω Ω– GNLS is BLUE in same cond s NLS, but Ω for I.– …don’t know Ω, so need consistent 1st stage (e.g., NLS)
• FGNLS is asymptotically BLUE:• FGNLS is asymptotically BLUE:2 2( , )+ with ( )
ˆ ˆ ˆf V σ σ= = ≠y X Iβ ε ε Ω
1 1 1
1 1 1 1 1 1
ˆ ˆ ˆ( )ˆ ˆ ˆ ˆ ˆ( ) ( ) ( ) ( )
FGNLS
V V
− − −
− − − − − −
′ ′⇒ = ∇ ∇ ∇′ ′ ′⇒ = ∇ ∇ ∇ ∇ ∇ ∇
y
y
β Ω Ωβ Ω Ω Ω Ω
1 1
( ) ( ) ( ) ( )ˆ( )
FGNLSV V
− −
⇒ ∇ ∇ ∇ ∇ ∇ ∇′= ∇ ∇
yβ Ω Ω Ω ΩΩ
Slide 12 of 37
Nonlinear Least-Squares:“Multiple Hands on the Wheel” Model (Franzese, PA ‘03)Multiple Hands on the Wheel Model (Franzese, PA 03)
• Monetary Policy in Open & Institutionalized EconK C&IPE I t /St t CBI ER R i M O– Key C&IPE Insts/Struct: CBI, ER-Regime, Mon. Open
• º CBI ≡ º Govt Delegated Mon Pol to CB• º Peg ≡ º Domestic (CB&Gov) Delegate to Peg-Curr (CB&Gov)eg o est c (C &Gov) e egate to eg Cu (C &Gov)• º FinOp ≡ º Dom cannot delegate b/c effectively del’d to globe
– Effect of ev’thing to which for. & dom. mon. pol-mkrs would d diff’l d d b i t t t & &respond diff’ly depends on combo insts-structs & v.v., &,
through intl inst-structs, for. on dom. & v.v.1 1 2 2( ) (1 ) ( )P E C P E Cπ π⋅ ⋅ ⋅ + ⋅ ⋅ − ⋅⎧ X X1 1 2 2
– Multicolinear Nightmare:• 23=8 inst-struct conds, i, times k factors per πi(Xi) if lin-interact
7 7 8 8(1 ) (1 ) ( ) (1 ) (1 ) (1 ) ( )P E C P E Cπ π⎪+ − ⋅ − ⋅ ⋅ + − ⋅ − ⋅ − ⋅⎩ X X
2 8 inst struct conds, i, times k factors per πi(Xi) if lin interact• Exponentially more if all polynominials; k!/2(k-2)! if all pairs.• Good thing can lean on some thry to specify more precisely!
Slide 13 of 37
Nonlinear Least-Squares:“Multiple Hands on the Wheel” ModelMultiple Hands on the Wheel Model
• CB & Govt Interaction (Franzese, AJPS ‘99):
c cπ π=
( ) ( ) (1 ) ( )π π π= ⋅ + − ⋅c c g gE c cx x
( ) ( , , , , , , , )π π π=g g g aGP UD BC TE EY FS AWx– Note: in this case, NL model nested within linear-
interactive model; test is that b bi/b equal ∀x.
c cπ π ( ) ( , , , , , , , )g g g a
interactive model; test is that bx⋅cbi/bx equal ∀x.
• Full Monetary Exposure & Atomistic ⇒ zero d ti tdomestic autonomy ⇒
• s.t. that, full e.r.fix⇒CB&Gov match peg⇒85 5 6 6
(1 ) (1 ) (1 ) (1 ) (1 ) ( )( ) ( ) π ππ π + − ⋅ − ⋅ ⋅ + − ⋅ − ⋅ − ⋅⎪= = ⎩⎪⎭ c gP E C P E C xx x
(1 )⎧E P E3 3 4 4
8
(1 )( ) ( )
(1 ) (1 ) (1 ) ( )
π ππ π π
π π
⋅ + ⋅ − ⋅⎧⎪= = ⇒ ⎨+ − ⋅ − ⋅ ⋅ + − ⋅⎡ ⎤⎪ ⎣ ⎦⎩
a p
pc g
E P E
P E C Cx x
xSlide 14 of 37
Nonlinear Least-Squares:“Multiple Hands on the Wheel” ModelMultiple Hands on the Wheel Model
• Compact & intuitive, yet gives all theoretically expected interactions, in the form expected
Slide 15 of 37
Nonlinear Least-Squares:“Multiple Hands on the Wheel” ModelMultiple Hands on the Wheel Model
• Effectively Estimable, yet gives all theoretically t d i t ti i th f t dexpected interactions, in the form expected
• Just 14 parameters (plus intercepts & dynamics,• Just 14 parameters (plus intercepts & dynamics, assuming those constant), just 3 more than lin-add!
• Parameters substantive meaning too:• Parameters substantive meaning, too:– Degree to which…constrains certain set of actors.
Yi ld t f i fl ti t t h th ti l f ll i d CB– Yields est. of inflation-target hypothetical fully indep CB• ⇒ general strategy for estimating/measuring unobservables:
– If know role factor will play & explanators of factor well enoughIf know role factor will play & explanators of factor well enough, can estimate unobservables conditional on both those theories, if both powerful enough & enough empirical variation.
Slide 16 of 37
Nonlinear Least-Squares:“Multiple Hands on the Wheel” ModelMultiple Hands on the Wheel Model
• Neat, but does it work? (Try it! Data online; stata: h l l) E ti t d E ti / Std Ehelp nl). Estimated Equation, w/ Std. Errs.:
• This project attempts a synthesis:Disting theoretically/conceptually many effects of #– Disting. theoretically/conceptually many effects of # (fragment.) & diversity (polar., partisan) policymakers.
– Empirical model of many effects distinctly & effectively.p y y y– Preliminary application to evolution fiscal policy (pub debt)
in developed democracies, 1950s-90s.Slide 21 of 37
• Tsebelis (‘95b, ‘99, ‘00, ‘02): Essential Argument:↑ # &/or ideological/interest polarization of pol mkng actors whose– ↑ # &/or ideological/interest polarization of pol-mkng actors whose approval required to ΔSQ, i.e., veto actors, ⇒, loosely, ↓ probability &/or magnitude policy Δ.I t i tl i W(SQ) ↓ hi h ll d # &/– I.e., strictly, as size W(SQ) ↓, which generally does as # &/or polarization VA ↑, range possible policy ∆(SQ) ↓.
– ⇒ following empirical prediction (Tsebelis 2002, Fig. 1.7):
– Suggests both mean/expected policy-∆ � & variance pol & pol-∆ ↑↓( ) ↑↓ ( )as size of W(SQ) ↑↓ (aside: why only suggests)
– No prediction of pol-level or of direction pol-∆, only of E(|∆p|), V(∆p).
Slide 22 of 37
Veto-Actor Implications↑ ( )• ↑ # (=Frag) & Polar of VA Privileges SQ ⇒– Retards policy-adjustment rates/delays stabilization,
• Results, e.g. in fiscal policy, deficits & debts; originally mixed, but tighter specify thry into empirical analysis:, g p y y p y– (F ’00, ’02) How model: policy-adjustment-rate effect =
conditional coefficient on LDV in dynamic model, not level.– (F ’00, ’02) How measure: frag & polar in VA theory =
• raw #, not eff. # (size-wtd) VA;
• max range pref’s, not V(pref’s) or sd(pref’s), (size-wtd)
distributive/pork-barrel spend (law of 1/n)Benefits concentrate district i: B =f(C) f’>0 & f”<0– Benefits concentrate district i: Bi=f(C), f >0 & f <0
– Costs disperse across n districts: Ci=C/n
⇒ ti l j t i f i’ i ↑ i # di t i t– ⇒optimal project-size from i’s view ↑ in # districts: f’(C*)=1/n (…log-linearly?)
• Alternative Decision Rules/Processes [ ] ⇒• Alternative Decision Rules/Processes […] ⇒– […] Law of 1/n is general, & stronger as legislative behavior more
Universalistic & less Minimal-Winning which tendency ↑ as rationalUniversalistic & less Minimal-Winning, which tendency ↑ as rational ignorance, winning-coalition uncertainty, or legislative-rule closure to amend or veto ↑.
– E.g., PubRev = common pool for n reps, overused to distribute bens; this CA prob worsens “proportionally” by law 1/n, i.e. at rate b/w those at which (n+1)/2n (MWC) & 1/n (uni) ↓ in nthose at which (n+1)/2n (MWC) & 1/n (uni) ↓ in n
Manifestations of Common Pools• Velasco (‘98, ‘99, ‘00): inter-temporal totality pub rev is C-P to
today’s policymakers ⇒ deficits & debts also law of 1/nP t & ’ T i f d li ⇒ lti l t• Peterson & co’s, Treisman: federalism ⇒ multiple tax authorities ⇒ several common-pool problems:– Inter-jurisdiction competition (w/ high factor mobility) ⇒ C-P of j p ( / g y)
investment resources ⇒ over-fishing: taxes too low.– National govt as lender last resort ⇒ subnational jurisdictions see fed
guarantee & funds as common pool ⇒ excessive borrowing by subnat’lg p g yunits. (e.g., EU, EMU & Euro ⇒ common pools…)
• Again, should be quite general:A thi th t i t f l k dit d i t d ↑– Anything that gains set of pol-makers credit ⇒ underinvested as ↑n
– Anything that gains set of pol-makers blame ⇒ overexploited as ↑n• E.g., (theory of the 2nd-best), ELECTIONEERING:g , (t eo y o t e best), C O R G
– Magnitude incentive electioneer fades w/ n (see, e.g., Goodhart)– Control over electioneering diminishes w/ n.
C A / & f f• Notice: CP not arise in Tsebelis’ VA Theory b/c # & pref’s of VA’s exog & predetermined, whereas in CP theory: prefs=f(#).
Slide 26 of 37
Modeling Common-Pool Effects• CP Effects distinguishable from VA Effects:
C P Eff l l ( i VA) i d i– C-P Effects on levels, not (as in VA) in dynamics.
– Proportional to 1/n for equal-sized actors. Standard Olsonian encompassingness logic ⇒ proper n here issize-weighted (effective & s.d./var.)
( ) ( )– Fractionalization (#) & esp. polarization (het.) relate to VA effects; CP, conversely, relate primarily to #, lth h h t b t CA balthough het. can exacerbate some CA probs.
• Suggests Proper Model of Policy-Response to some public demand for, x1'β1, or against, x2'β2:– …+(x1'β1)(1-f(ln(Eff#))+(x2'β2)(1+f(ln(Eff#))+…( 1 β1)( f( ( ff#)) ( 2 β2)( f( ( ff#))– Same f(ln(Eff#), b/c overexploit/underinvest same º
huge theoretical & empirical literatureg• F (‘99,‘02,‘03): less context-specific empirical strategy…
– Because broad comparativist seek thry that travels, not that requires different model each contextrequires different model each context.
• Offering is roughly equivalent Nash Bargaining.– Most ext forms ⇒ eqbm bounded by actors’ ideal pts:q y p
• Convex set/hull, upper-contour set (=core of coop. game thry),• So like Tsebelis, but further, though short of explicit ext-form
– Policy outside that range possiblePolicy outside that range possible,• e.g., if uncertainty resolved unfavorably,• but that ⇒ highly unlikely that E(pol) outside this range
Th E( l) b ( td ) l k ’– Thus, E(pol)=some convex-combo (wtd-avg) pol-mkrs’ ideals ⇒ convex-combo emp. models ≈ compromise
• If Nash Bargain, e.g., (n.b. NB=coop. game-thry but equiv. sev. )reasonable ext-form non-coop barg. games: Rubinstein ‘82), ⇒
Empirical Manifestations & Modelof Compromise Policymakingp y g
• Re: def’s & debt, Cusack (‘99, ‘01; cf., Clark ‘03)– Arg: left more Keynes-active counter-cyc; right less, even pro-cyc
– Add Nash-Barg Model ⇒ wtd-avg pol-mkr partisanship conditions º Keynesian cntr-cyc fisc-pol response to macroecon.
• Empirical Implementation:– Ideally:
• Describe barg power party i as f(charact’s i & barg envir, j, ⇒ f(vij)• Desc pol response to conditions x if i sole pol mkng control: q (x )• Desc. pol response to conditions xk if i sole pol-mkng control: qi(xk)
• Then embed Nash-Barg sol’n, Σif(vij)qi(xk), in emp. model to est.
– D-B Effects: power-wtd mean ideologies (partisanship)
• Different ways these distinct effects manifest in pol:y p– V-A (primarily) to slow pol-adjust (delay stabilization);
– C-P induces over-draw from common resources (incl. from (future as in debt); under-invest in common properties (less electioneering), log-proportionately
t li l/ ti li l i ti tcountercyclical/conservative pro-cyclical, in proportion to degree left/right controls policymaking
Slide 30 of 37
Empirical Model of the Theoretical Synthesis (2)
• …implies specification where:– Abs # VA & ideol range modify pol-adjust ratesAbs # VA & ideol range modify pol adjust rates
– (log) Eff # pol-mkrs & s.d. ideol (wtd measures) gauge C P prob in electioneering (+debt lvl effect?)gauge C-P prob in electioneering (+debt-lvl effect?)
– Some barg process among partisan pol-mkrs (e.g., N h td i fl ) d t i b fl t dNash ⇒ wtd-influence) determines combo reflected in net policy responsiveness to macro (º K-activism)
• ⇒( ) ( )1 1 2 2 3 31it i it it i t i t i tD NoP ARwiG D D Dα ρ ρ ρ ρ ρ= + + + × + +( ) ( )
( ) ( )1 , 1 2 , 2 3 , 3
, , ,
1
1
it i n it ar it i t i t i t
Y i t U i t P i t cg it
D NoP ARwiG D D D
Y U P CoG
α ρ ρ ρ ρ ρ
β β β β− − −
Δ Δ Δ
+ + + × + +
+ Δ + Δ + Δ × +
( ) ( )1 2 , 1 1e it e i t en it sd it it it itE E ENoP SDwiGγ γ γ γ ε−+ + × + + + + +′ ′x η z ω
Slide 31 of 37
Empirical Model Specification & Data( ) ( )1D NoP ARwiG D D Dα ρ ρ ρ ρ ρ ε+ + + × + + + + +′ ′x η z ω
b (%G )
( ) ( )( ) ( ) ( ) ( )
1 , 1 2 , 2 3 , 3
, , , 1 2 , 1
1
1 1
it i n it ar it i t i t i t it it it
Y i t U i t P i t cg it e it e i t en it sd it
D NoP ARwiG D D D
Y U P CoG E E ENoP SDwiG
α ρ ρ ρ ρ ρ ε
β β β β γ γ γ γ− − −
Δ Δ Δ −
= + + + × + + + + +
+ Δ + Δ + Δ × + + + × + +
x η z ω
Dit = Debt (%GDP);NoP & ARwiG = raw Num of Prtys in Govt & Abs Range w/i Govt:
VA conception so modify dynamics Expect ρ & ρ >0VA conception, so modify dynamics. Expect ρn & ρar >0.By thry & for efficiency: modify all lag dynamics same.
CoG (govt center, left to right, 0-10):Modifies response to macroecon (equally, by thry & for eff’cy) : βcg<0.Macroec: ΔY = real GDP growth; ΔU = Δ unemp rate; ΔP = infl rate.
x’η = controls: set pol econ cond’s response to which not partisanx η = controls: set pol-econ cond s response to which not partisan-differentiated or gov comm-pool: (e.g., E(real-int)-E(real-grow), ToT)
ENoP & SDwiG = Effective Num of Prtys in govt & Std Dev w/i Govt:F & P l b d fl CP l l ff d f (Frag & Polar by wtd-influence concept. CP lvl-effects modify (at same
rate) electioneering, pre-elect: Et & post-elect: Et-1: γen & γsd<0.z’ω = set of constituent terms in the interactions:
ENoP, SDwiG may have positive coeff’s by CP effect lvl debt, but issue is temporal fract, not curr. govt fract. Thry o/w says omit.
• Joint-significance of multiple-policymaker conditioning effects (γen, γsd, ρn, ρar, βcg) overwhelmingly rejects excluding (p≈.001), whereas joint-sig coeff’s on constit. terms, z, clearly fails reject (p≈.602) exclusion. (Almost) All h h ld b (SD iG & AR iG l h & )theory says should be zero (SDwiG & ARwiG closest thry & emp.), so…
Coeff. Std. Err. t-Stat. Pr(T>|t|)Dt-1 1.207 0.060 20.290 0.000 D
Lagged D d Dt-2 -0.158 0.085 -1.851 0.065Dependent Variables Dt-3 -0.117 0.045 -2.577 0.010
• x0 = factors that affect policy-outcomes unless pol-mkrs act to change status quo, i.e., that have effect on pol-out directly.
• x1 = factors affecting policy-outcomes if policymakers act to g p y f p ychange status quo, without partisan-differentiated response
• x2 = factors affecting policy-outcomes if policymakers act to change status quo with partisan differentiated responsechange status quo, with partisan-differentiated response
• {NoP,ARwiG} = sources of veto-actor effects; as before• {ENoP,SDwiG} = sources of common-pool effects; as before{ , } p ;• {p(cit),qj(xt)} = sources of bargaining & delegation effects:
– p(cit): Effective policy-influence of party i in context t. (E.g., as now: cabinet seat-shares but could become richer model )cabinet seat shares, but could become richer model.)
– qj(xt): Model of response of party i to pol-econ conditions xt. (E.g., as now: CoGi×Macroecont, but could become richer model.)
Slide 36 of 37
Preliminary Results of Fuller Model Coeff. Std. Err. t-Stat. Pr(T>|t|)( | |)