Scribd
Upload
42 views6 pages

Multivariate Correlational Research Insights

Beta is a standardized regression coefficient that indicates the strength and direction of the relationship between a predictor variable and an outcome variable, while controlling for other predictor variables. A third variable problem occurs when a predictor appears to have a significant relationship with an outcome in a bivariate correlation but becomes insignificant when accounting for other variables in multiple regression. Adding more predictor variables to a regression analysis can help rule out potential third variable relationships and increase internal validity, but regression does not prove causation.

Uploaded by

Kristen Starks
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
42 views6 pages

Multivariate Correlational Research Insights

Beta is a standardized regression coefficient that indicates the strength and direction of the relationship between a predictor variable and an outcome variable, while controlling for other predictor variables. A third variable problem occurs when a predictor appears to have a significant relationship with an outcome in a bivariate correlation but becomes insignificant when accounting for other variables in multiple regression. Adding more predictor variables to a regression analysis can help rule out potential third variable relationships and increase internal validity, but regression does not prove causation.

Uploaded by

Kristen Starks
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

Chapter 9 Notes: Multivariate Correlational Research

BETA = standardized regression coefficient; how we understand how different factors affect our
outcome variable

 NOT on a -1 to 1 scale like r


 (+) and (–) denote direction of association (like r)
 Units are standard deviations
o Going up by 1 SD in a certain predictor, β = how many SDs does outcome variable
changes, controlling all the other factors in the analysis
o Ex: F or age, 0.33 – for a 1 SD unit increase in age,
pregnancy risk goes up by 0.33 SDs
o Even a really small value of β can be very
relevant, if significant and much higher than the
other predictors
 Statistical significance:
o If beta is not significant (p > .05), we say there is
no relationship b/t predictor and the DV when
other predictors are controlled for
o If it is significant (p < .05), we say there is a true
relationship b/t the predictor and the DV when the other predictors are controlled for
o p-values = the likelihood of getting stats of that magnitude (when controlling for other
variables) just by random chance if the null hypothesis were true

THIRD VARIABLE PROBLEM occurs when a predictor has a significant relationship with an outcome in a
bivariate correlation, but then becomes insignificant when you account for other variables in a multiple
regression

Adding More Predictors to a Regression

 Point – rule out more potential 3rd variable


relationships -> increase internal validity
 Betas change as you put more variables in and
take others out
 For example: could also add,
o Parental supervision
o age
o interest in sex
o social relationships/life
o religiosity
o location
o SES
o Access to healthcare
o contraception
o free time
o personality
 Still cannot claim causation because not every single possible factor has been accounted for
 βs for dichotomous variables (coded as 0 or 1) may tend to be larger because of the inherently
lower amount of spread in the data b/c there are only 2 options. So, changing from one option
to another is a big change in standard deviation

Beta (β) is different from b:

 Both are indicators of the relationship between predictor and outcome variables
 b is not standardized, based on the same units as whatever the original scale was
o cannot compare b values b/c they are on diff scales
 For ex: a 1-unit change on a 5-point scale is diff. than a 1-unit change on a 9-
point scale.
o can be easier to interpret based on the actual variable meaning
 β is standardized using standard deviation, so you can compare betas

Regression in Popular Press Articles:

 Often use phrases such as to imply that multiple regression was used:
o “controlled for”
o taking into account
o

Regression does not establish causation:

 multiple regression is NOT a foolproof way to rule out all kinds of third variables

Getting at Causality with Pattern and Parsimony:

 When you can’t do an experiment because of ethics or feasibility, you need lots and lots of
consistent studies
 the longer a person has smoked cigarettes, the greater the risk of cancer

Pattern, Parsimony, and the Popular Press

 Journalists do not always fairly represent pattern and parsimony


 When journalists report only one study at a time, they are selectively presenting only part of the
scientific process.
Types of Third Variables

MEDIATION – A “why?” problem

 The mechanism behind the relationship between two variables


 A hypothesized causal sequence

Examples:

Being teased about weight -> Weight gain

Emotional eating?
Fear of Participating in team
sports → low activity?
Depression?

Being teased about Weight Gain


your weight

Grades - > Happiness

Parental Approval
Self-esteem

Grades Happiness

Basics of Mediation Analysis

 Requires at least 3 time points


o To assess temporal precedence/directionality:
o If you measure all the variables at the same time, you don’t know which order they
occur in
 Mediated effects are referred to as an INDIRECT EFFECTS
o Grades have an “indirect effect” on happiness through self-esteem
o Being teased about your weight has an “indirect effect” on weight gain through
decreased physical activity
 Kids might not have decreased their physical activity, leading to weight gain, if
the teasing had not occurred in the first place
 What you look for statistically:
o Covariance
o 4 steps:
 Does X predict Y?
 Does X predict M?
 Does M predict Y?
 When X and M are entered together in a multiple regression, does M still
predict Y, and does X no longer predict Y?

Grades no longer sig.


X Happiness Y

M
Self-Esteem
sig sig

no longer sig. Happiness


X Grades Y

Mediators vs. Third Variables

THIRD VARIABLES

 External effect
 A and B are only related because C predicts A and C predicts B
 Ex: Ice cream and drowning

MEDIATORS

 Internal effect

Mediators vs. Moderators

 Similarities:
o Both analyses involve a third variable that has some sort of effect on the relationship
o ____
 Mediation: Why?
o Internal Validity issue
 Moderator: For WHOM and WHEN?
o External validity issue
o Can be different strengths, or the relationship can exist for one group and not exist for
the other
 Mediators and Moderators can go together
o MODERATED MEDIATION
o Does the mediation relationship differ based on levels of yet another (4th) variable?

Practice: Mediator, Moderator, or Third Variable?

a. Having a mentally demanding job is associated with cognitive benefits in later years
because people who are highly educated take mentally demanding jobs, and people
who are highly educated have better cognitive skills.

Third Variable

b. Having a mentally demanding job is associated with cognitive benefits in later years, but
only in men, not women.

Moderator

c. Having a mentally demanding job is associated with cognitive benefits in later years
because cognitive challenges build lasting connections in the brain.

Mediator

d. Being a victim of sibling aggression is associated with poor mental health in childhood,
but the link is especially strong for later-born children and weaker in firstborn children.

Moderator

e. Sibling aggression is associated with poor childhood mental health because child victims
of sibling aggression are more likely to feel lonely at home. Sibling aggression leads to
loneliness, which in turn leads to mental health problems.

Mediator

f. Sibling aggression is associated with poor childhood mental health only because of
parental conflict. Sibling aggression is more likely to occur when parents argue
frequently, and arguing also affects kids’ mental health.
Third Variable

Common questions

Powered by AI

Researchers cannot establish causation from regression results alone because regression, a key tool in multivariate correlational research, addresses statistical relationships, not causative links. It lacks insights on temporal precedence and can neither ensure all relevant variables are accounted for nor exclude omitted variable bias. Hence, regression results remain correlational, not causal, regardless of finding significant β values .

A significant beta (β) value in a regression analysis indicates a true relationship between the predictor and the dependent variable, notwithstanding the control for other predictors in the model. This significance is determined by a p-value less than 0.05, implying that the results are unlikely to have occurred due to random chance under a true null hypothesis .

Mediation explains the 'why?' behind a relationship, acting as an indirect effect where one variable influences a second variable through a mediator (internal effect). In contrast, a third variable is an external factor that provides an alternate explanation for a relationship between two variables; in this case, both variables are correlated only because they independently associate with the third variable .

Regression analysis cannot establish causation because it only shows association, not the direction of the relationship or any temporal precedence. Controlling for third variables does not eliminate all possible confounding factors, and regression models rely on observed data which might not capture underlying causal mechanisms .

When journalists report single studies selectively, they risk misrepresenting the scientific process by omitting the necessity of pattern and parsimony, which require consistent results across multiple studies for reliable conclusions. This selective reporting can lead the public to misunderstand research as isolated events rather than as parts of a cumulative process, resulting in skewed perceptions about the reliability and applicability of scientific findings .

Moderated mediation is useful in research as it explores whether the mediating effect of a variable differs across levels of another variable, effectively answering 'For whom and when?' this mediation is stronger or weaker. This approach provides insights into conditional processes, showcasing the nuanced complexities in relationships between variables, which can alleviate ambiguity in broader applications of the mediation findings .

An indirect effect in mediation analysis is confirmed when the predictor (X) influences an outcome (Y) through a mediator (M). Statistically, this is demonstrated by four steps: X significantly predicts M; M predicts Y; and when X and M are included in a regression model together, M still predicts Y while X no longer does. This sequence supports the mediator's role in the relationship .

Establishing temporal precedence in mediation analysis is crucial because it confirms the directionality of the relationships between the variables. This process requires capturing data at least at three different time points to ensure that the presumed mediator occurs after the predictor but before the outcome variable. Without this, the supposed causal order could be questioned, weakening the mediation analysis's validity .

Adding more predictors in regression analysis can enhance internal validity by ruling out additional potential third-variable relationships. This approach narrows down the causes that might influence the dependent variable, thus increasing the likelihood that any observed association is not due to uncontrolled confounding variables .

β values may tend to be larger for dichotomous variables because of the inherently lower variance compared to continuous variables. Since dichotomous variables have only two levels, any change represents a significant shift in the standard deviation, leading to a pronounced effect in the regression coefficient in relation to other predictor variables .

You might also like