Knight Foundation Internal Analysis

STEP 1: Recodings and data prep

Most steps taken in this stage of the analysis are data-cleaning related, and difficult to show in this format. However, it is important to document the steps. They include:

Original data file had a value code of -98. This was set to missing, meaning respondents coded this way for any particular question will be exluded from the base for any calculations related to that question (this extends further into multivariate analysis)
Items Q6-Q14 presented respondents with two statements and asked them to describe which more closely aligned with their own views. In determing how best to handle these questions, it became apparent that these questions can be treated as basically agree/disagree questions. All items were recoded so that the low-end of the scale (i.e. responses 1 or 2) signify a person is expressing a position that is anti-regulation/anti-government intervention/anti-censorship, depending on the context of the question. Responses at the 4 or 5 point of the scale are expressing the opposite.
Other items in the Q3 series (trust in democracy) and Q4 series (statements about society and politics) were recoded to be in a similar ‘direction’ (i.e. a similar point of view). This was done primarily to make factor analysis results more easily intepretable.

STEP 2: Factor Analysis: Q3 Items

We first turn to the Q3 Items to determine a potential composite variable, and the sub-indices that may be included. Most graphics in the below output will refer to the variable names, so it is worth repeating them here:

Q3AR:The government provides aid for those who need it.
Q3BR:The government promotes a free and open market.
Q3CR:People choose their leaders in free elections.
Q3DR:All adult citizens enjoy the same legal and political rights.
Q3ER:People are allowed to express political views freely.
Q3FR:The government protects individuals’ right to engage in peaceful protest.
Q3GR:News organizations are free from government inference.

We first look at the overall results of these items.

dep_var	Wording	1 - Not at all essential	2	3	4	5 - Absolutely essential
Q3A	The government provides aid for those who need it.	5.2	11.7	25.2	23.0	34.9
Q3B	The government promotes a free and open market.	4.0	7.2	25.5	26.3	37.0
Q3C	People choose their leaders in free elections.	1.8	1.9	5.5	7.5	83.2
Q3D	All adult citizens enjoy the same legal and political rights.	3.7	3.5	6.8	10.0	76.0
Q3E	People are allowed to express political views freely.	2.0	2.7	8.6	19.8	66.9
Q3F	The government protects individuals’ right to engage in peaceful protest.	2.6	3.4	10.2	16.1	67.7
Q3G	News organizations are free from government inference.	4.8	4.7	14.3	16.9	59.3

Next, we look at the correlation between these items. Given these items are on a 5-point scale, we are using polychoric correlations. However, the idea is the same – correlations closer to 1 indicate a very tight association. However, this is a higher bar for polychoric collerations. In general a correlation of around 0.4 for this type of analysis is fairly strong.

q3.data<-survey.df %>%
  dplyr::select(Q3AR:Q3GR, WEIGHT)%>%
  drop_na(.)

observations<-round(sum(q3.data[,"WEIGHT"]),0)

q3.polycor<-psych::polychoric(q3.data[1:7], weight=q3.data$WEIGHT, na.rm=TRUE)

##All respondents

cor.plot(q3.polycor$rho, numbers=T, upper=TRUE, main = "Q3 Series Polychoric Correlation\nAll respondents", show.legend = FALSE)

While the correlations are generally strong, it is notable that Q3A (Government provides aid for those who need it) does not fit especially strongly with any of the other question items.

From here, we start factor analysis. Below plots show the scree plot, which helps to determine the number of factors. Depending on the method used, we see somewhere between 2-3 suggested components. We default to the principal axis factor analysis, which suggests 3 factors.

###FACTOR ANALYSIS


fa.parallel(q3.polycor$rho, fm="pa", fa="both",n.obs=observations, main = "Q3 Items: Scree Plot")

## Parallel analysis suggests that the number of factors =  3  and the number of components =  2

###3 factor

q3_model<-fa(q3.data[1:7], nfactor=3, cor="poly", fm="mle", rotate = "varimax", weight=q3.data$WEIGHT)
q3_model$loadings

## 
## Loadings:
##      ML3    ML1    ML2   
## Q3AR  0.119         0.876
## Q3BR  0.422  0.456       
## Q3CR  0.498  0.811  0.299
## Q3DR  0.666  0.453  0.335
## Q3ER  0.793  0.369       
## Q3FR  0.763  0.289  0.355
## Q3GR  0.643  0.364  0.140
## 
##                  ML3   ML1   ML2
## SS loadings    2.508 1.429 1.120
## Proportion Var 0.358 0.204 0.160
## Cumulative Var 0.358 0.562 0.723

fa.diagram(q3_model, main="Factor Analysis with 3 factors: Q3 Items")

Notably, Q3A is its own factor. This suggests trying a 2 factor solution.

q3_model_2<-fa(q3.data[1:7], nfactor=2, cor="poly", fm="mle", rotate = "varimax", weight=q3.data$WEIGHT)
q3_model_2$loadings

## 
## Loadings:
##      ML2    ML1   
## Q3AR  0.121  0.990
## Q3BR  0.606       
## Q3CR  0.834  0.281
## Q3DR  0.827  0.309
## Q3ER  0.851       
## Q3FR  0.790  0.330
## Q3GR  0.742  0.132
## 
##                  ML2   ML1
## SS loadings    3.661 1.285
## Proportion Var 0.523 0.184
## Cumulative Var 0.523 0.707

fa.diagram(q3_model_2, main="Factor Analysis with 2 factors: Q3 Items")

These results further suggest we should consider treating these items as one scale, rather than multiple indicators. As a final test, we examine the Cronbach’s Alpha score.

cronbach.alpha(q3.data[1:7], CI=TRUE, na.rm=TRUE)

## 
## Cronbach's alpha for the 'q3.data[1:7]' data-set
## 
## Items: 7
## Sample units: 10085
## alpha: 0.817
## 
## Bootstrap 95% CI based on 1000 samples
##  2.5% 97.5% 
## 0.806 0.826

We find the score to be alpha=0.817 which suggests this can be treated as its own scale.

CLUSTER ITEMS

The main question items of interest for the purposes of cluster analysis are Q5-Q14, or

Q5: Regulation of social media – companies or government
Q6: Responsibility of social media posts – individual or social media companies
Q7: Responsibilty for identifying false information on social media – individual or social media companies
Q8: Limits of posting and sharing content on social media – none or social media companies should
Q9: Does censorship justify stopping fake news – no to yes
Q10: Expressing political views on social media: Has no impact to does have impact
Q11: Proving identity to create social media account: Does not need to or should
Q12: Worry about information on social media – more worried about censorship or misleading information
Q13: Being anonymous on social media – allows free expression to making it easier to post harmful information
Q14: Tech companies take positions with respect to politics – they should not do so or should do so

All items have been rescaled so that a LOW score on the five point scale indicates the person preferred less action OR advocated for the “lower level” concept mentioned in the question (for instance in the first question about whether regulation should be implemented by the company or the government, the “lower level” entity would be the company) OR generally preferred unrestricted freedom OR preferred a position typically advocated by conservative politicans.

This analysis was conducted in two different ways – one using variables which had simple “agree,” or “not agree” categories for each statement. In this formulation, there were 20 variables included in the cluster analysis.

The alternative approach was to use the full 5-point scale variables across the 10 questions. This ultimately was the settled upon approach, though the reasons why will not be explored here.

Cluster Analysis: Initial Data Checks

var1	var2	correlation	pval	abs_corr
Q8R	Q12R	0.6391693	0.0000000	0.6391693
Q7R	Q9R	0.5536667	0.0000000	0.5536667
Q6R	Q8R	0.5442790	0.0000000	0.5442790
Q7R	Q12R	0.5248919	0.0000000	0.5248919
Q6R	Q12R	0.5225662	0.0000000	0.5225662
Q9R	Q12R	0.4947119	0.0000000	0.4947119
Q12R	Q13R	0.4426996	0.0000000	0.4426996
Q7R	Q8R	0.4365979	0.0000000	0.4365979
Q8R	Q13R	0.4285354	0.0000000	0.4285354
Q12R	Q14R	0.4121119	0.0000000	0.4121119
Q11R	Q13R	0.3839941	0.0000000	0.3839941
Q6R	Q7R	0.3578773	0.0000000	0.3578773
Q8R	Q9R	0.3461516	0.0000000	0.3461516
Q8R	Q14R	0.3382148	0.0000000	0.3382148
Q5R	Q6R	0.3156127	0.0000000	0.3156127
Q7R	Q14R	0.3065021	0.0000000	0.3065021
Q6R	Q14R	0.3063516	0.0000000	0.3063516
Q6R	Q13R	0.3020753	0.0000000	0.3020753
Q6R	Q9R	0.2981618	0.0000000	0.2981618
Q9R	Q10R	0.2418834	0.0000000	0.2418834
Q7R	Q10R	0.2416778	0.0000000	0.2416778
Q9R	Q14R	0.2344481	0.0000000	0.2344481
Q5R	Q8R	0.2236245	0.0000000	0.2236245
Q9R	Q11R	0.2156407	0.0000000	0.2156407
Q7R	Q13R	0.2039738	0.0000000	0.2039738
Q7R	Q11R	0.2004104	0.0000000	0.2004104
Q5R	Q13R	0.1811737	0.0000000	0.1811737
Q10R	Q11R	0.1767256	0.0000000	0.1767256
Q9R	Q13R	0.1744242	0.0000000	0.1744242
Q5R	Q12R	0.1690545	0.0000000	0.1690545
Q8R	Q11R	0.1528869	0.0000000	0.1528869
Q13R	Q14R	0.1519079	0.0000000	0.1519079
Q10R	Q12R	0.1313499	0.0000000	0.1313499
Q6R	Q10R	0.1278842	0.0000000	0.1278842
Q8R	Q10R	0.1219126	0.0000000	0.1219126
Q11R	Q12R	0.1167931	0.0000000	0.1167931
Q5R	Q11R	0.1126229	0.0000000	0.1126229
Q6R	Q11R	0.1004600	0.0000000	0.1004600
Q11R	Q14R	-0.0736169	0.0000000	0.0736169
Q5R	Q10R	0.0701165	0.0000000	0.0701165
Q10R	Q14R	0.0578486	0.0000000	0.0578486
Q5R	Q9R	0.0517552	0.0000003	0.0517552
Q10R	Q13R	0.0463883	0.0000042	0.0463883
Q5R	Q7R	0.0384960	0.0001488	0.0384960
Q5R	Q14R	0.0128358	0.2038356	0.0128358

We see the question items are relatively poorly correlated – the highest being Q8R/Q12R at 0.64.

This suggests factor analysis (PCA) for the purposes of data reduction is not necessary and we can proceed into cluster analysis.

We will be performing hierarchical clustering. Given the ordinal nature of the data, dissimilarity is calculated using the “gower” metric, which is available in the “cluster” package of R.

cluster.data[,cluster.items1]<-lapply(cluster.data[,cluster.items1], factor)

class(survey.df$Q5_BR)

## [1] "numeric"

cluster.data[,cluster.items]<-lapply(cluster.data[,cluster.items], as.ordered)

####SHOWING SILHOUTTES

hclust_optimizer(cluster.data[,cluster.items])

This shows that are best number of clusters (in terms of internal consistency) is essentially 2 – otherwise, as the number of clusters grow, this measure becomes increasingly weaker.

There is an exception to this pattern, however – when the number of clusters is equal to 6. As can be seen in the graph, the internal consistency improves slightly (even if it is well below the smaller cluster numbers). Given our desire for a greater number of clusters, this seems like a good number to try.

We now look at our initial clusters and overall variable importance.

###Plotting cluster solution

hclust_plotter(cluster.data[,cluster.items], 6)

gower.pp.core <- daisy(cluster.data[, cluster.items], metric=c("gower"))
hc <- hclust(gower.pp.core, method='ward.D2')
groups <- cutree(hc, k=6)
disc<-rf_discriminator(cluster.data[, cluster.items], groups)

##      MeanDecreaseGini
## Q5R          373.7077
## Q6R          574.0728
## Q7R         1061.1090
## Q8R          795.2104
## Q9R         1032.3889
## Q10R         381.7179
## Q11R         542.0356
## Q12R        1190.0025
## Q13R         641.2825
## Q14R         509.7625

disc_matrix<-as.data.frame(disc, row.names = NULL)

disc_var<-rownames(disc_matrix)

gini_matrix<-cbind(disc_matrix,disc_var)

gini_matrix%>%
  arrange(desc(MeanDecreaseGini))%>%
  left_join(label_df, by=c("disc_var"="QTAG"))%>%
  dplyr::select(disc_var,Wording,MeanDecreaseGini)%>%
  kbl() %>%
  kable_styling()

disc_var	Wording	MeanDecreaseGini
Q12R	Worry about information on social media – more worried about censorship or misleading information	1190.0025
Q7R	Responsibilty for identifying false information on social media – individual or social media companies	1061.1090
Q9R	Does censorship justify stopping fake news – no to yes	1032.3889
Q8R	Limits of posting and sharing content on social media – none or social media companies should	795.2104
Q13R	Being anonymous on social media – allows free expression to making it easier to post harmful information	641.2825
Q6R	Responsibility of social media posts – individual or social media companies	574.0728
Q11R	Proving identity to create social media account: Does not need to or should	542.0356
Q14R	Tech companies take positions – they should not do so or should do so	509.7625
Q10R	Expressing political views on social media: Has no impact to does have impact	381.7179
Q5R	Regulation of social media – companies or government	373.7077

The above table shows the relative importance of each variable in the cluster analysis. It is clear that the questions about censorship vs misleasing information, responsibility for identifying false information and the role of censorship in preventing fake news are the items that bear the greatest signficance in how our clusters are formed.

Note, however, that this analysis does not suggest any question should be removed from the cluster analysis, and we proceed ahead – specifying 6 groups.

label_alt <- as.factor(cutree(hc, k=6))
disc2<-rf_discriminator(cluster.data[, cluster.items], label_alt)

##      MeanDecreaseGini
## Q5R          375.3508
## Q6R          584.4130
## Q7R         1063.0247
## Q8R          798.3157
## Q9R         1033.4358
## Q10R         380.6900
## Q11R         543.9977
## Q12R        1158.4212
## Q13R         648.6543
## Q14R         515.6952

cluster.data$label_alt <- as.character(cutree(hc, k=6))

table(cluster.data$label_alt)

## 
##    1    2    3    4    5    6 
## 1566 2779  836 1863 1748  394

cluster.toplines<-pollster::topline(cluster.data, label_alt, WEIGHT)

cluster.toplines%>%
   kbl() %>%
  kable_styling()

Response	Frequency	Percent	Valid Percent	Cumulative Percent
1	1791.4855	19.157394	19.157394	19.15739
2	2760.8965	29.523868	29.523868	48.68126
3	862.1214	9.219165	9.219165	57.90043
4	1807.6020	19.329736	19.329736	77.23016
5	1755.0932	18.768230	18.768230	95.99839
6	374.2066	4.001608	4.001608	100.00000

cluster.toplines$Cluster<-factor(cluster.toplines$Response, levels=c(1,2,3,4,5,6),
                                 labels=c("One", "Two", "Three", "Four", "Five", "Six"))

lattice::barchart(Cluster ~ Percent, data=cluster.toplines,
                  main="Overall Cluster Results")

We have six groups, with cluster 2 representing the largest share of the population (29.5%). Cluster 6 is the smallest, at 4%.

NOTE: 9% OF RESPONDENTS ARE EXCLUDED FROM THIS ANALYSIS DUE TO MISSING DATA

How do the clusters compared on key items?

cluster.results2<-survey_two_tab_looper(cluster.data, "label_alt", cluster.items1)



cluster.results2$cluster.group<-factor(cluster.results2$ind_var, 
                                      levels=c(1,2,3,4,5,6),
                                      labels=c("One", "Two", "Three", "Four", "Five", "Six"))

cluster.wide2<-cluster.results2%>%
  dplyr::filter(dep_category == "1")%>%
  dplyr::select(cluster.group,dep_var,Wording,pct)%>%
  mutate(pct = round(pct,1))%>%
  dplyr::select(cluster.group,dep_var,Wording,pct)%>%
  pivot_wider(names_from=cluster.group, values_from=pct)


cluster.wide2%>%
  kbl() %>%
  kable_styling()

dep_var	Wording	One	Two	Three	Four	Five	Six
Q5_AR	Social media companies should make their own policies without regulation about what people can post	47.2	24.7	48.8	34.3	43.0	15.8
Q5_BR	The government should regulate and enforce the way social media companies take down false or harmful content	16.1	59.4	24.3	36.8	34.0	51.4
Q6_AR	Social media users alone should be responsible for the content they post online	47.0	12.3	71.8	39.0	85.9	8.0
Q6_BR	Social media companies shoul dbe responsible for the content their users post online	21.3	76.1	13.0	42.9	8.0	75.5
Q7_AR	Social media companies should identify information when it is false or misleading	59.5	96.7	58.4	57.0	5.9	4.0
Q7_BR	Individual users should be responsible for fact-checking content	17.3	0.8	24.0	32.1	87.3	84.5
Q8_AR	People should have the freedome to say whatever they want online	34.3	1.8	77.8	17.1	80.9	15.2
Q8_BR	Social media companies should limit abusive or threatening language online	33.7	93.8	8.2	73.0	12.6	65.9
Q9_AR	Fake news is a more serious problem than censorship online	51.7	90.6	85.8	70.7	7.3	9.1
Q9_BR	Censorship online is a more serious problem than fake news	16.0	3.3	2.7	10.6	83.8	76.6
Q10_AR	Expressing political views on social media has a significant impact on politics	49.1	75.1	65.7	54.4	48.7	15.0
Q10_BR	Expressing political views on social media does not have any real world impact on politics	13.0	10.0	10.2	23.7	31.3	60.6
Q11_AR	People should have to provide proof of their identity when they create accounts on social media	28.0	68.3	72.9	66.7	45.5	7.7
Q11B_R	People should not have to provde who they are when they create accounts on social media	34.0	13.0	6.8	9.4	37.3	60.8
Q12_AR	I am worried about information online being removed or censored	20.9	1.2	74.1	9.3	91.1	25.9
Q12_BR	I am worried about false and misleading information online	38.9	94.4	3.1	77.0	2.9	49.3
Q13_AR	Being anonymous online allows people to freely express themselves on social media	25.9	2.4	55.8	3.7	37.5	20.4
Q13_BR	Being anonymous online makes it easier to post harmful or abusive content and bully people on social media	30.3	90.4	19.2	88.2	44.9	56.5
Q14_AR	Major technology companies should not take a position on political and social issues	34.2	29.1	79.2	48.7	85.0	29.8
Q14_BR	Major technology companies should be able to take a position on political and social issues	27.9	46.3	4.3	25.4	6.3	43.8

Examination of key items using the full scale.

Demographics of cluster groups.

Cluster	18 TO 34	35 TO 54	55+
One	39.9	32.8	27.3
Two	36.1	31.2	32.7
Three	17.9	38.1	44.1
Four	26.5	34.5	39.0
Five	28.7	36.3	35.0
Six	19.9	31.5	48.6

Cluster	COLLEGE GRADUATE	NON-COLLEGE GRADUATE
One	34.9	65.1
Two	52.6	47.4
Three	19.4	80.6
Four	35.5	64.5
Five	25.3	74.7
Six	22.4	77.6

Cluster Analysis: Alternative Solution

An alternative solution focuses on using a binary version of these variables – for any given statement a person can agree or not agree (including those who choose option 3). For this analysis, we will now have 20 variables as there will be one variable per statement.

As we can see in the below Silhoutte Width plot, the optimal solution appears to be 4.

##        MeanDecreaseGini
## Q5_AR          87.46580
## Q5_BR          96.51225
## Q6_AR         217.62811
## Q6_BR         146.55189
## Q7_AR         550.80852
## Q7_BR         440.91386
## Q8_AR         571.93393
## Q8_BR         610.12607
## Q9_AR         270.74414
## Q9_BR         251.54274
## Q10_AR        111.50800
## Q10_BR         70.77965
## Q11_AR        104.56550
## Q11B_R         69.97108
## Q12_AR        656.23658
## Q12_BR        693.91593
## Q13_AR        112.57576
## Q13_BR        166.91193
## Q14_AR        150.17696
## Q14_BR         69.21450

## [1] "QTAG"    "Wording"

disc_var	Wording	MeanDecreaseGini
Q12_BR	I am worried about false and misleading information online	693.91593
Q12_AR	I am worried about information online being removed or censored	656.23658
Q8_BR	Social media companies should limit abusive or threatening language online	610.12607
Q8_AR	People should have the freedome to say whatever they want online	571.93393
Q7_AR	Social media companies should identify information when it is false or misleading	550.80852
Q7_BR	Individual users should be responsible for fact-checking content	440.91386
Q9_AR	Fake news is a more serious problem than censorship online	270.74414
Q9_BR	Censorship online is a more serious problem than fake news	251.54274
Q6_AR	Social media users alone should be responsible for the content they post online	217.62811
Q13_BR	Being anonymous online makes it easier to post harmful or abusive content and bully people on social media	166.91193
Q14_AR	Major technology companies should not take a position on political and social issues	150.17696
Q6_BR	Social media companies shoul dbe responsible for the content their users post online	146.55189
Q13_AR	Being anonymous online allows people to freely express themselves on social media	112.57576
Q10_AR	Expressing political views on social media has a significant impact on politics	111.50800
Q11_AR	People should have to provide proof of their identity when they create accounts on social media	104.56550
Q5_BR	The government should regulate and enforce the way social media companies take down false or harmful content	96.51225
Q5_AR	Social media companies should make their own policies without regulation about what people can post	87.46580
Q10_BR	Expressing political views on social media does not have any real world impact on politics	70.77965
Q11B_R	People should not have to provde who they are when they create accounts on social media	69.97108
Q14_BR	Major technology companies should be able to take a position on political and social issues	69.21450

##        MeanDecreaseGini
## Q5_AR          89.15638
## Q5_BR          96.16201
## Q6_AR         221.95715
## Q6_BR         135.58362
## Q7_AR         537.82387
## Q7_BR         443.15320
## Q8_AR         592.35260
## Q8_BR         569.23201
## Q9_AR         264.19819
## Q9_BR         266.49364
## Q10_AR        110.03150
## Q10_BR         69.84877
## Q11_AR        103.74671
## Q11B_R         70.41488
## Q12_AR        686.84534
## Q12_BR        686.79002
## Q13_AR        110.48912
## Q13_BR        177.16160
## Q14_AR        151.34005
## Q14_BR         69.61862

##        MeanDecreaseGini
## Q5_AR          89.15638
## Q5_BR          96.16201
## Q6_AR         221.95715
## Q6_BR         135.58362
## Q7_AR         537.82387
## Q7_BR         443.15320
## Q8_AR         592.35260
## Q8_BR         569.23201
## Q9_AR         264.19819
## Q9_BR         266.49364
## Q10_AR        110.03150
## Q10_BR         69.84877
## Q11_AR        103.74671
## Q11B_R         70.41488
## Q12_AR        686.84534
## Q12_BR        686.79002
## Q13_AR        110.48912
## Q13_BR        177.16160
## Q14_AR        151.34005
## Q14_BR         69.61862

Results for Cluster Analysis Solution 2

This shows selexted results for Cluster Analysis Solution 2, including overall frequencies and how the clusters responded to the key cluster question items.

Cluster Solution 2 Overall Results
Response	Frequency	Percent	Valid Percent	Cumulative Percent
1	4525.201	48.39060	48.39060	48.39060
2	1390.529	14.86973	14.86973	63.26032
3	1023.280	10.94252	10.94252	74.20285
4	2412.396	25.79715	25.79715	100.00000

Results of Cluster Solution 2 by Cluster Items
dep_var	Wording	One	Two	Three	Four
Q5_AR	Social media companies should make their own policies without regulation about what people can post	30.2	59.3	15.2	42.9
Q5_BR	The government should regulate and enforce the way social media companies take down false or harmful content	50.6	14.4	28.4	33.8
Q6_AR	Social media users alone should be responsible for the content they post online	20.9	67.1	20.2	81.3
Q6_BR	Social media companies shoul dbe responsible for the content their users post online	64.6	14.5	41.0	10.3
Q7_AR	Social media companies should identify information when it is false or misleading	84.6	84.9	20.2	7.2
Q7_BR	Individual users should be responsible for fact-checking content	10.2	3.9	46.3	81.9
Q8_AR	People should have the freedome to say whatever they want online	3.1	74.9	13.6	74.7
Q8_BR	Social media companies should limit abusive or threatening language online	90.1	8.1	43.3	17.3
Q9_AR	Fake news is a more serious problem than censorship online	84.1	77.9	16.2	22.8
Q9_BR	Censorship online is a more serious problem than fake news	5.2	8.7	41.0	65.3
Q10_AR	Expressing political views on social media has a significant impact on politics	68.9	63.4	21.5	49.4
Q10_BR	Expressing political views on social media does not have any real world impact on politics	13.3	12.0	29.4	30.3
Q11_AR	People should have to provide proof of their identity when they create accounts on social media	62.5	58.1	23.0	48.9
Q11B_R	People should not have to provde who they are when they create accounts on social media	17.3	16.3	29.4	31.9
Q12_AR	I am worried about information online being removed or censored	2.4	39.8	15.3	86.8
Q12_BR	I am worried about false and misleading information online	90.1	33.5	33.4	3.1
Q13_AR	Being anonymous online allows people to freely express themselves on social media	5.5	46.4	11.4	33.1
Q13_BR	Being anonymous online makes it easier to post harmful or abusive content and bully people on social media	83.9	30.3	44.2	46.8
Q14_AR	Major technology companies should not take a position on political and social issues	34.8	56.7	22.7	82.3
Q14_BR	Major technology companies should be able to take a position on political and social issues	40.1	18.6	27.3	8.1