slides_Ch4_W[1]

# 4 but rst why would we care about such an hypothesis

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: s We are now ready to learn how regression output can be used to test simple hypothesis of interest involving the βj , for instance Ho : β j = 0 Melissa Tartari (Yale) Econometrics (4.4) 19 / 41 A Simple Hypothesis We are now ready to learn how regression output can be used to test simple hypothesis of interest involving the βj , for instance Ho : β j = 0 (4.4) But …rst, why would we care about such an hypothesis and what is it that we are testing? Melissa Tartari (Yale) Econometrics 19 / 41 A Simple Hypothesis We are now ready to learn how regression output can be used to test simple hypothesis of interest involving the βj , for instance Ho : β j = 0 (4.4) But …rst, why would we care about such an hypothesis and what is it that we are testing? Because βj measures the partial e¤ect of xj on the expected value of y after controlling for all other indep. vars, (4.4) means that, once x1 , ...xj 1 , xj +1 , ..., xK have been accounted for, xj has no e¤ect on E [y jx] i.e. (4.4) is equivalent to Ho : Melissa Tartari (Yale) ϑ E [y jx] =0 ϑxj Econometrics 19 / 41 A Simple Hypothesis - Example Let log wage = βo + β1 educ + β2 exp + β3 exp2 + β4 tenure + u Melissa Tartari (Yale) Econometrics 20 / 41 A Simple Hypothesis - Example Let log wage = βo + β1 educ + β2 exp + β3 exp2 + β4 tenure + u A simple hypothesis of interest could be: Ho : β 4 = 0 Melissa Tartari (Yale) Econometrics 20 / 41 The t-Test We will see that the statistics that we use to test (4.4) is tβ ˆ tβ = ˆ j ˆ β hj i ˆ se β jx j (4.5) j Melissa Tartari (Yale) Econometrics 21 / 41 The t-Test We will see that the statistics that we use to test (4.4) is tβ ˆ tβ = ˆ j ˆ β hj i ˆ se β jx j (4.5) j QUESTION: why is tβ “reasonable” as a test to detect βj 6= 0? ˆ j Melissa Tartari (Yale) Econometrics 21 / 41 The t-Test i h ˆ ˆ Recall: tβ = βj /se βj jx ˆ j Melissa Tartari (Yale) Econometrics 22 / 41 The t-Test i h ˆ ˆ Recall: tβ = βj /se βj jx ˆ j Observe: Melissa Tartari (Yale) Econometrics 22 / 41 The t-Test i h ˆ ˆ Recall: tβ = βj /se βj jx ˆ j Observe: ˆ βj is an unbiased estimator of βj so it is only natural that we look at if for guidance Melissa Tartari (Yale) Econometrics 22 / 41 The t-Test i h ˆ ˆ Recall: tβ = βj /se βj jx ˆ j Observe: ˆ βj is an unbiased estimator of βj so it is only natural that we look at if for guidance i h hi hi ˆ ˆ since se βj jx > 0 always, ) sign tβ = sign βj ˆ j Melissa Tartari (Yale) Econometrics 22 / 41 The t-Test i h ˆ ˆ Recall: tβ = βj /se βj jx ˆ j Observe: ˆ βj is an unbiased estimator of βj so it is only natural that we look at if for guidance i h hi hi ˆ ˆ since se βj jx > 0 always, ) sign tβ = sign βj ˆ j h i ˆ ˆ for a given value of se βj jx , a larger βj ) larger tβ ˆ j Melissa Tartari (Yale) Econometrics 22 / 41 The t-Test i h ˆ ˆ Recall: tβ = βj /se βj jx ˆ j Observe: ˆ βj is an unbiased estimator of βj so it is only natural that we look at if for guidance i h hi hi ˆ ˆ since se βj jx > 0 always, ) sign tβ = sign βj ˆ j h i ˆ ˆ for a given value of se βj jx , a larger βj ) larger tβ ˆ j ˆ Hence, a sample value of βj very di¤erent from 0 provides evidence against Ho : βj = 0 ; however we must recognize that there is ˆ ˆ sampling error in our estimate βj , so the deviation of βj from 0 must h i ˆ ˆ be weighted against the sampling error of βj . Because se βj jx is an h i ˆ estimate of sd βj jx , tβ measures how many estimated standard ˆ j ˆ is away from 0. deviations β j Melissa Tartari (Yale) Econometrics 22 / 41 The t-Test: Digression The formal rendition of this intuitive answer is that the test is unbiased. Let us recap on terminology: a test is a statistics, a statistics is a function (solely) of the sample data (prior to the sampling process), hence it is a random variable and as such it has a distribution (which does not depend on unknown parameters). An unbiased test is a random variable whose distribution under Ho is centered at zero. For instance, E [tβ jHo : βj = 0] = 0, and similarly we will see that ˆ j n o E [FQ jHo : βj = 0jj 2 Q ] = 0. Melissa Tartari (Yale) Econometrics 23 / 41 2- Sided Alternative Consider the case in which Ho : βj = 0 is tested against H1 : βj 6= 0. Melissa Tartari (Yale) Econometrics 24 / 41 2- Sided Alternative Consider the case in which Ho : βj = 0 is tested against H1 : βj 6= 0. Let tβ be the value of the statistics computed using sample ˆ j information. Melissa Tartari (Yale) Econometrics 24 / 41 2- Sided Alternative Consider the case in which Ho : βj = 0 is tested against H1 : βj 6= 0. Let tβ be the value of the statistics computed using sample ˆ j information. First, we choose a signi…cance level α (probability of rejecting Ho when it is true): e.g. α = 5%. Melissa Tartari (Yale) Econometrics 24 / 41 2- Sided Alternative Consider the case in which Ho : βj = 0 is tested against H1 : βj 6= 0. Let tβ be the value of the statistics computed using sample ˆ j information. First, we choose a signi…cance level α (probability of rejecting Ho when it is true): e.g. α = 5%. Then, you can proceed in one of two ways: Melissa Tartari (Yale) Econometrics 24 / 41 2- Sided Alternative Consider the case in which Ho : βj = 0 is tested against H1 : βj 6= 0. Let tβ be the value of the statistics computed using sample ˆ j information. First, we choose a signi…cance level α (probability of rejectin...
View Full Document

## This note was uploaded on 02/13/2014 for the course ECON 350 taught by Professor Donaldbrown during the Fall '10 term at Yale.

Ask a homework question - tutors are online