Participants (n = 957) were asked how strongly they agree with the following statements.
In the first lines of code below, I import and clean the data. Most importantly, I removed anyone who said they weren’t a parent and don’t plan on becomeing one in the near future.
# Import, clean
dat=read.csv("/Users/marklacour/Desktop/Junk/Google Drive/Research/--Research -- Advance/0Parents/Wave 1 summary/Capps SEPA/wave1.csv")
dat=subset(dat,dat$Status != "Survey Preview") # Don't keep any "Survey Preview" rows
dat=dat[-c(1:2),]
dat=subset(dat,dat$switchboard != "I am not a parent, and don’t plan on becoming one")After that, I recode the data to turn the verbal (“Strongly agree”, “Somewhat disagree”, etc.) data into numerical data.
######################################################
#### Parent status
parent.status=ifelse(dat$switchboard=="I am a parent, but my child (or children) have moved out of the house","past",NA)
parent.status=ifelse(dat$switchboard=="I am a parent, currently raising a child (or children)","present",parent.status)
parent.status=ifelse(dat$switchboard=="I plan on becoming a parent in the near future","future",parent.status)
dat$parent.status=parent.status
######################################################
#### online medical information seeking
seeking.future=cbind(
as.numeric(car::recode(dat$seeking1.future, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking2.future, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking5.future, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking6.future, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7"))
)
seeking.present=cbind(
as.numeric(car::recode(dat$seeking1.present, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking2.present, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking5.present, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking6.present, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7"))
)
seeking.past=cbind(
as.numeric(car::recode(dat$seeking1.past, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking2.past, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking5.past, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7")),
as.numeric(car::recode(dat$seeking6.past, "'Strongly disagree'=1; 'Disagree'=2; 'Somewhat disagree'=3; 'Neither agree nor disagree'=4; 'Somewhat agree'=5; 'Agree'=6; 'Strongly agree'=7"))
)
# combind across parenting statuses
a=seeking.past
a=ifelse(is.na(a),seeking.present,a)
a=ifelse(is.na(a),seeking.future,a)
online.health.info=rowMeans(seeking.past)
online.health.info=ifelse(is.na(online.health.info),rowMeans(seeking.present),online.health.info)
online.health.info=ifelse(is.na(online.health.info),rowMeans(seeking.future),online.health.info)
dat$online.health.info=online.health.infoThere is no simple, bivariate correlation between age and the degree to which parents report seeking medical information online.
dat$age=as.numeric(dat$age)
mod=lm(dat$online.health.info~dat$age)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$age)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0080 -1.0077 0.2421 0.9927 2.9929
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.008e+00 1.560e-01 25.697 <2e-16 ***
## dat$age -1.964e-05 3.865e-03 -0.005 0.996
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.441 on 952 degrees of freedom
## (3 observations deleted due to missingness)
## Multiple R-squared: 2.713e-08, Adjusted R-squared: -0.00105
## F-statistic: 2.583e-05 on 1 and 952 DF, p-value: 0.9959When I observe null effects, I always like to examine the (approximate) Bayes factor comparing the null hypothesis to the alternative hypothesis (\(BF_{01}\)).
null.mod=lm(dat$online.health.info~1)
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 4.944109The null hypothesis (that age isn’t associated with internet information seeking) is almost 5 times more likely than the alternative. This is fairly strong evidence favoring the null, but it’s not decisive.
There is no association between gender and tendency to seek medical information online.
dat$is.male=ifelse(dat$gender=="Male",1,0)
dat$is.third=ifelse(dat$gender!="Male" & dat$gender!="Female",1,0)
mod=lm(dat$online.health.info~dat$is.male)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$is.male)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0866 -0.9583 0.1634 1.0417 3.0417
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.95833 0.05878 67.337 <2e-16 ***
## dat$is.male 0.12829 0.09641 1.331 0.184
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.44 on 953 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 0.001854, Adjusted R-squared: 0.0008069
## F-statistic: 1.77 on 1 and 953 DF, p-value: 0.1836And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 12.73831Yup, looks like gender (by itself, taking no other variables into consideration) is definitely not related to online information seeking.
There is no statistically significant difference in online information seeking between heterosexual participants and LGBTQ+ ones.
dat$is.bi=ifelse(dat$orientation=="Bisexual",1,0)
dat$is.homosexual=ifelse(dat$orientation=="Homosexual (Gay, Lesbian)",1,0)
dat$is.lgbt=ifelse(dat$orientation!="Heterosexual (Straight)",1,0)
mod=lm(dat$online.health.info~dat$is.lgbt)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$is.lgbt)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0157 -1.0157 0.2343 1.0430 3.0430
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.01566 0.05101 78.718 <2e-16 ***
## dat$is.lgbt -0.05866 0.12582 -0.466 0.641
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.441 on 953 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 0.000228, Adjusted R-squared: -0.0008211
## F-statistic: 0.2174 on 1 and 953 DF, p-value: 0.6412And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 27.71461That’s a big Bayes Factor. The null hypothesis wins this one big time.
There is no statistically significant relationship between white and non-white participants.
is.white=ifelse(dat$race=="White",1,0)
dat$is.white=is.white
is.asian=ifelse(dat$race=="Asian",1,0)
dat$is.asian=is.asian
is.black=ifelse(dat$race=="Black or African American",1,0)
dat$is.black=is.black
is.multiple=ifelse(grepl(",",dat$race),1,0)
dat$is.multiple=is.multiple
mod=lm(dat$online.health.info~dat$is.white)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$is.white)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.023 -1.023 0.227 1.027 3.027
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.97308 0.07993 49.704 <2e-16 ***
## dat$is.white 0.04994 0.09842 0.507 0.612
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.441 on 953 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 0.0002701, Adjusted R-squared: -0.0007789
## F-statistic: 0.2575 on 1 and 953 DF, p-value: 0.612And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 27.16316Another big win for the null hypothesis.
No statistically significant difference betwen Hispanic and non-Hispanic participants.
is.latino=ifelse(dat$latino=="Hispanic or Latino or Spanish Origin",1,0)
dat$is.latino=is.latino
mod=lm(dat$online.health.info~dat$is.latino)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$is.latino)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0149 -1.0149 0.2351 0.9851 3.0714
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.01488 0.04922 81.564 <2e-16 ***
## dat$is.latino -0.08631 0.15366 -0.562 0.574
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.441 on 953 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 0.0003309, Adjusted R-squared: -0.000718
## F-statistic: 0.3155 on 1 and 953 DF, p-value: 0.5745And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 26.3855Big win for the null
Nothing for English as first langauge.
# English first language
english.not.first=ifelse(dat$language != "Yes",1,0)
dat$english.not.first=english.not.first
mod=lm(dat$online.health.info~dat$english.not.first)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$english.not.first)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0077 -1.0077 0.2423 0.9923 3.0298
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.00767 0.04770 84.023 <2e-16 ***
## dat$english.not.first -0.03743 0.22744 -0.165 0.869
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.441 on 953 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 2.842e-05, Adjusted R-squared: -0.001021
## F-statistic: 0.02708 on 1 and 953 DF, p-value: 0.8693And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 30.48658The null dominates.
No statistically significant association between political affiliation and online information seeking
# Politics
pols=dat$politics
pols=ifelse(pols=="Very liberal",1,pols)
pols=ifelse(pols=="Somewhat liberal",2,pols)
pols=ifelse(pols=="Neutral",3,pols)
pols=ifelse(pols=="Somewhat conservative",4,pols)
pols=ifelse(pols=="Very conservative",5,pols)
pols=as.numeric(pols)
dat$pols=pols
mod=lm(dat$online.health.info~dat$pols)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$pols)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0340 -1.0088 0.2286 1.0165 3.0165
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.97086 0.11963 33.194 <2e-16 ***
## dat$pols 0.01264 0.03984 0.317 0.751
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.442 on 950 degrees of freedom
## (5 observations deleted due to missingness)
## Multiple R-squared: 0.0001058, Adjusted R-squared: -0.0009467
## F-statistic: 0.1006 on 1 and 950 DF, p-value: 0.7512And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 0.2481476The null wins again.
There IS a statistically significant relationship between income (considered by itself) and online information seeking. Namely, people of higher income tend to seek more information online.
income=dat$income
income=ifelse(income=="$0 - $29,999",1,income)
income=ifelse(income=="$30,000 - $59,999",2,income)
income=ifelse(income=="$60,000 - $89,999",3,income)
income=ifelse(income=="$90,000 - $119,999",4,income)
income=ifelse(income=="$120,000 or more",5,income)
income=as.numeric(income)
dat$income=income
mod=lm(dat$online.health.info~dat$income)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$income)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.2505 -0.9975 0.2495 1.1260 3.2554
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.61810 0.11348 31.884 < 2e-16 ***
## dat$income 0.12648 0.03367 3.756 0.000183 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.432 on 950 degrees of freedom
## (5 observations deleted due to missingness)
## Multiple R-squared: 0.01464, Adjusted R-squared: 0.0136
## F-statistic: 14.11 on 1 and 950 DF, p-value: 0.0001828People with higher education levels are significantly more likely to seek out health information online.
education=dat$Education
education=ifelse(education=="Some high school",1,education)
education=ifelse(education=="High school diploma, or equivalent",2,education)
education=ifelse(education=="Some college",3,education)
education=ifelse(education=="Bachelor's Degree",4,education)
education=ifelse(education=="Master's Degree",5,education)
education=ifelse(education=="Doctorate or other advanced degree (e.g., M.D., J.D.)",6,education)
dat$education=as.numeric(education)
mod=lm(dat$online.health.info~dat$education)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$education)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.3831 -1.0541 0.1959 1.1104 3.1104
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.39618 0.17819 19.059 < 2e-16 ***
## dat$education 0.16448 0.04634 3.549 0.000405 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.432 on 951 degrees of freedom
## (4 observations deleted due to missingness)
## Multiple R-squared: 0.01307, Adjusted R-squared: 0.01204
## F-statistic: 12.6 on 1 and 951 DF, p-value: 0.000405People who practice a religion and people who don’t (e.g., atheists, agnostics) are no different in terms of how much they report seeking out information online.
# religion (category)
not.religious=ifelse(dat$religion=="Not religious (e.g., Atheist, Agnostic)",1,0)
dat$not.religious=not.religious
mod=lm(dat$online.health.info~dat$not.religious)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$not.religious)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0486 -0.9814 0.2014 1.0186 3.0186
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.98140 0.05858 67.966 <2e-16 ***
## dat$not.religious 0.06717 0.09676 0.694 0.488
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.441 on 953 degrees of freedom
## (2 observations deleted due to missingness)
## Multiple R-squared: 0.0005053, Adjusted R-squared: -0.0005435
## F-statistic: 0.4818 on 1 and 953 DF, p-value: 0.4878And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 24.27643The null dominates.
The degree to which someone considers themselves “religious” does not correlate with oneline information seeking.
# religiosity (continuum)
religious=dat$religious
religious=ifelse(religious=="Strongly disagree",1,religious)
religious=ifelse(religious=="Disagree",2,religious)
religious=ifelse(religious=="Somewhat disagree",3,religious)
religious=ifelse(religious=="Neither agree nor disagree",4,religious)
religious=ifelse(religious=="Somewhat agree",5,religious)
religious=ifelse(religious=="Agree",6,religious)
religious=ifelse(religious=="Strongly agree",7,religious)
dat$religiosity=as.numeric(religious)
mod=lm(dat$online.health.info~dat$religiosity)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$religiosity)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0524 -1.0062 0.1976 1.0316 3.0484
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.06919 0.09240 44.040 <2e-16 ***
## dat$religiosity -0.01680 0.02165 -0.776 0.438
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.442 on 950 degrees of freedom
## (5 observations deleted due to missingness)
## Multiple R-squared: 0.0006333, Adjusted R-squared: -0.0004187
## F-statistic: 0.602 on 1 and 950 DF, p-value: 0.438And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 0.2596109Somehow, we got a non-significant result but \(BF_{01}\) indicates the null is less likely than the alternative. Maybe I made a mistake in my code? If not, that’s an odd result.
There’s no significant correlation between how “spiritual” someone is and their online information seeking habits.
# Spirituality
spiritual=dat$spiritual
spiritual=ifelse(spiritual=="Strongly disagree",1,spiritual)
spiritual=ifelse(spiritual=="Disagree",2,spiritual)
spiritual=ifelse(spiritual=="Somewhat disagree",3,spiritual)
spiritual=ifelse(spiritual=="Neither agree nor disagree",4,spiritual)
spiritual=ifelse(spiritual=="Somewhat agree",5,spiritual)
spiritual=ifelse(spiritual=="Agree",6,spiritual)
spiritual=ifelse(spiritual=="Strongly agree",7,spiritual)
dat$spiritual=as.numeric(spiritual)
mod=lm(dat$online.health.info~dat$spiritual)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$spiritual)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0190 -1.0138 0.2362 1.0072 2.9967
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.982348 0.120189 33.134 <2e-16 ***
## dat$spiritual 0.005235 0.024217 0.216 0.829
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.442 on 952 degrees of freedom
## (3 observations deleted due to missingness)
## Multiple R-squared: 4.908e-05, Adjusted R-squared: -0.001001
## F-statistic: 0.04673 on 1 and 952 DF, p-value: 0.8289And let’s see the \(BF_{01}\)…
bf01=exp((BIC(mod)-BIC(null.mod))/2)
bf01## [1] 8.218665For once the null isn’t that much more likely than the alternative. This still seems like a pretty solid win for the null though.
While many of the simple, bivariate relationships were not statistically significant, the analyses were also very simplistic on this initial run. Namely, I placed most diverse categories into a binary: “part of the numerical majority” versus “NOT part of the numerical majority.” This resulted in simplistic analyses like “white participants compared to non-white participants” and “hetero participants compared to LGBTQ+ participants.
The problem with looking at a bunch of individual pairs of variables is that each pair only considers the two variables in isolation. With multiple regression, we are seeing whether the dependent variable (online information seeking) can be predicted from a 2+ predictor variables. When you take multiple predictor variables into account, the inter-correlations between each one is taken into account. In other words, maybe age (considered by itself) does not correlate with online information seeking. But maybe when age and income are both in the model together you can see that age (while adjusting for income) DOES predict online information seeking.
In the model below, I didn’t put EVERY demographic as a predictor. There are trade-offs; more predictors are better, but too many predictors (compared to how much data you have) can be a problem.
mod=lm(dat$online.health.info~
dat$age+dat$is.male+dat$is.lgbt+dat$is.white+dat$pols+dat$income+dat$education+dat$religiosity)
summary(mod)##
## Call:
## lm(formula = dat$online.health.info ~ dat$age + dat$is.male +
## dat$is.lgbt + dat$is.white + dat$pols + dat$income + dat$education +
## dat$religiosity)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.5179 -1.0050 0.1722 1.0962 3.5402
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.224611 0.266749 12.089 <2e-16 ***
## dat$age -0.001120 0.003954 -0.283 0.7771
## dat$is.male 0.095314 0.096885 0.984 0.3255
## dat$is.lgbt 0.030353 0.131138 0.231 0.8170
## dat$is.white 0.045669 0.101321 0.451 0.6523
## dat$pols 0.042168 0.044920 0.939 0.3481
## dat$income 0.090335 0.037159 2.431 0.0152 *
## dat$education 0.130898 0.052329 2.501 0.0125 *
## dat$religiosity -0.034547 0.024025 -1.438 0.1508
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.431 on 941 degrees of freedom
## (7 observations deleted due to missingness)
## Multiple R-squared: 0.02412, Adjusted R-squared: 0.01582
## F-statistic: 2.907 on 8 and 941 DF, p-value: 0.003305It looks like income and education still have “unique” contributions to predicting online information seeking, but including all these variables together in the same model didn’t help highlight the unique contributions of other variables. In other words, the other, non-significant variables still don’t appear to be predictive of online information seeking, even after you take into account the inter-correlations AMONG these predictor variables.
Another strategy might be to look at interactions. For instance, we could ask, “Do men and women of different income levels seek information online differently from one another?” This isn’t the same as having gender and income in the same model.
It might be worthwhile to break people down by their parenting status: “Not yet, but soon”, “raising kids right now”, and “raised kids in the past but they’ve moved out.” The analyses above lumps all these people together.