Good point Michael: that's a great paper!
Omitted variables can include ordering. Note that if one can find
"instruments"--exogenous variables that are uncorrelated with the error
terms--to predict one or both of the endogenous variables then the
locked-in causal direction of the relationship (as estimated by 2SLS or
ML emulating 2SLS) should be the same whether one or the other construct
were to be measured first.
Best regards,
John
__________________________________________
Prof. John Antonakis
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis
Associate Editor
The Leadership Quarterly
__________________________________________
On 13.06.2011 17:51, Michael Johnson wrote:
> A further complication that is often not considered (and also doesn't answer
> Avi's request for data) is the order in which people respond to these
> scales. Schwarz (1999) noted that in a study where people were asked about
> how satisfied they were with their life as a whole and then asked how
> satisfied they were with their marriage, the correlation between the two was
> .32. When the order was reversed, however, the correlation was .67.
>
> His explanation is that when people are asked about specific satisfactions
> first, they assimilate these judgments into their overall satisfaction, but
> when they are asked about overall satisfaction first, they decompose this
> into specific satisfactions. I would guess that in most cases where
> supervisor satisfaction appears before general job satisfaction on the
> survey, the correlation would be inflated relative to the reverse ordering.
>
> Schwarz, N. (1999). Self-reports: How the questions shape the answers.
> American Psychologist, 54, 93-105.
>
> Michael D. Johnson
> Assistant Professor of Management and Organization
> Foster School of Business
> University of Washington
> Box 353200
> Seattle, WA 98195
> (206) 616-2756
>
mdj3@uw.edu
>
>
>
> -----Original Message-----
> From: Organizational Behavior Division Listserv
> [mailto:
OB@AOMLISTS.PACE.EDU] On Behalf Of John Antonakis
> Sent: Sunday, June 12, 2011 11:56 PM
> To:
OB@AOMLISTS.PACE.EDU
> Subject: Re: [OB-LIST] satisfaction with one's supervisor and general job
> satisfaction
>
> Hi Avi:
>
> I agree with Steve on this point, which I would like to extend. I know
> this is not answering your request for data and it a methodological
> point; however, it might be a beneficial discussion for those following
> the thread.
>
> Both satisfaction with one's supervisor (S) and general job satisfaction
> (J) are endogenous. Many models suffer the same problem of endogeneity
> (e.g., leader-member exchange used as predictor of outcomes). By
> endogenous, I means that the modeled independent variable is caused by
> something, and if this something is not fully measured and included in
> the model then estimates that are reported will be spurious. This
> something could be:
>
> 1. omitted (common) causes (e.g., A follower high on conscientiousness
> might be more satisfied with their work; also, because the follower is
> conscientious, the supervisor uses less punitive leadership styles with
> the worker. Thus, satisfaction with the supervisor and job satisfaction
> both depend partly on follower conscientiousness. This is just an
> example of an omitted common cause, but thing get more complex, as
> indicated below).
>
> 2. common method variance (similar to the above problem).
>
> 3. simultaneity (i.e., S-->J and J-->S)
>
> The correlation r(sj), is thus confounded with this "something." To
> understand this something (e and u below), let's express this relation
> in regression form.
>
> Eq 1: S = b0 + b1J + e
> Eq 2: J = g0 + g1S + u
>
> Substituting Eq. 2 into Eq. 1 gives (we can do it the other way round
> too) and ignoring the constant:
>
> S = b0 + b1(g0 + g1S + u) + e
>
> As you can see b1 will be affected by g1 and u, which are actually
> pooled in the error term (c) when simply estimating S = f0 + f1S + c.
> Thus, S will correlate with the disturbance, which violates the
> assumption of the regression (OLS) estimator (or ML estimator); whatever
> estimates are obtained cannot be trusted. For example, if J goes up for
> some unknown reason (u), it will affect S in Eq. 1; thus, what relation
> is captured in b1 is not at all clear. Just estimating Eq. 1 assumes J
> is exogenous and b1 captures the effect of J on S but this relation
> could be due to u too or to S affecting J.
>
> So, suppose one estimates r(sj) to be .30. I often see researchers
> saying something like: "due to the correlational nature of the data,
> this relation cannot be interpreted causally. S could be causing J, but
> J could be causing S. Thus, the coefficient is capturing the relation or
> association between these two variables" In fact, this reasoning is
> completely wrong. Shying away from using causal language is not helpful
> to understanding the true relation between the variables. And, the true
> relation between S and J includes many "relations"; when these relations
> are fully modeled the residual relation between S and J could actually
> be anything but .30 (i.e., higher, lower, negative, nil). Also, the
> effect of S on J and the effect of J on S is not necessarily the same.
>
> To get to the true relation between S and J, one would need to model:
>
> S = b0 + b1J + b2Z + v
> J = g0 + g1S + g2Z + w
>
> Where Z is a vector of omitted common causes. Of course, the above
> assumes that you know what Z is. One way to get around not knowing Z is
> to find "instruments" that is, exogenous variables that are unrelated to
> the error terms, but which predict the endogenous variables. Then one
> could estimate the following and "lock-in" the causal direction:
>
> S = m0 + m1J + k
> J = n0 + n1X + q
>
> X is the instrument. If it strongly predicts J and also predicts S less
> strongly, AND is, theoretically, unrelated to k and q, then m1 can be
> estimated consistently (i.e., one obtains the true parameter as the
> sample size increases). Ideally, one has more instruments than
> endogenous regressors (overidentified). We talk about this modeling
> strategy in depth (and a lot more) in the following paper for those who
> are interested:
>
> Antonakis, J., Bendahan, S., Jacquart, P.,& Lalive, R. (2010). On
> making causal claims: A review and recommendations. The Leadership
> Quarterly, 21(6). 1086-1120. Available here:
>
http://www.hec.unil.ch/jantonakis/Causal_Claims.pdf
>
> Here's the abstract:
> Social scientists often estimate models from correlational data, where
> the independent variable has not been exogenously manipulated; they also
> make implicit or explicit causal claims based on these models. When can
> these claims be made? We answer this question by first discussing design
> and estimation conditions under which model estimates can be
> interpreted, using the randomized experiment as the gold standard. We
> show how endogeneity - which includes omitted variables, omitted
> selection, simultaneity, common-method variance, and measurement error -
> renders estimates causally uninterpretable. Second, we present methods
> that allow researchers to test causal claims in situations where
> randomization is not possible or when causal interpretation could be
> confounded; these methods include fixed-effects panel, sample selection,
> instrumental variable, regression discontinuity, and
> difference-in-differences models. Third, we take stock of the
> methodological rigor with which causal claims are being made in a social
> sciences discipline by reviewing a representative sample of 110 articles
> on leadership published in the previous 10 years in top-tier journals.
> Our key finding is that researchers fail to address at least 66% and up
> to 90% of design and estimation conditions that make causal claims
> invalid. We conclude by offering 10 suggestions on how to improve
> non-experimental research.
>
> Best regards,
> John.
>
> __________________________________________
>
> Prof. John Antonakis
> Faculty of Business and Economics
> Department of Organizational Behavior
> University of Lausanne
> Internef #618
> CH-1015 Lausanne-Dorigny
> Switzerland
> Tel ++41 (0)21 692-3438
> Fax ++41 (0)21 692-3305
>
http://www.hec.unil.ch/people/jantonakis
>
> Associate Editor
> The Leadership Quarterly
> __________________________________________
>
>
> On 13.06.2011 01:44, Kelman, Steven wrote:
>> When you look at any studies, be careful about common method bias!
>>
>>
>> Steve Kelman
>> Albert J. Weatherhead III and Richard W.
>> Weatherhead Professor of Public Management
>> Editor, International Public Management Journal
>> Tel: 617-496-6302
>> Personal Homepage:
>>
http://www.ksg.harvard.edu/fs/skelman
>> IPMJ Homepage:
>>
http://www.tandf.co.uk/journals/titles/10967494.asp
>> "The Lectern," my blog on FCW. com
>>
http://fcw.com/blogs/lectern/list/blog-list.aspx
>>
>> -----Original Message-----
>> From: Organizational Behavior Division Listserv
> [mailto:
OB@AOMLISTS.PACE.EDU] On Behalf Of Avi Kluger
>> Sent: Sunday, June 12, 2011 2:52 AM
>> To:
OB@AOMLISTS.PACE.EDU
>> Subject: [OB-LIST] satisfaction with one's supervisor and general job
> satisfaction
>> Dear OBNet readers:
>>
>> I am looking for published papers correlating satisfaction with one's
> supervisor (boss) and general job satisfaction. Any lead would be greatly
> appreciated
>> Avi Kluger