8 Sensitivity analyses

8.1 Sensitivity analysis 1 : Refitting primary model but with a CS structure

# Sensitivity analysis 1 : with a CS structure
Meta.S1<-rma.mv(yi,V.Prim, 
                  mods=~Predictor-1, 
                  random = ~ Out.Pred | Study, 
                  struct="CS", 
                  data=ES.Prim, level=100-(5/4))

df.Meta.Res1.S1<-as.data.frame(ExtractMetaFUNCTION(Meta.S1))

df.Meta.S1<- df.Meta.Res1.S1 %>% 
  dplyr::mutate(
    Study="Pooled Effect Size",
    Outcome="All outcomes",
    SleepIndicator=c("Quality of Acute Sleep", "Quality of Chronic Sleep",
                     "Quantity of Acute Sleep","Quantity of Chronic Sleep"),
    Adj.p.val=case_when(
      Adj.p.val>.99~.99,
      Adj.p.val<=.99~Adj.p.val),
    N=sum(N.prim)) %>%
  rowwise() %>%
  dplyr::mutate(TOST=max(TOST1, TOST2)) %>%
  dplyr::select(-c(TOST1,TOST2))

df.Meta.S1<-df.Meta.S1[,c(8:11,1:7,12)]


#Summary of the meta analysis
gt(df.Meta.S1)%>% 
  fmt_number(
    columns = 5:12,
    decimals = 3)%>%
  cols_merge(
    columns=vars(Adj.CIlow, Adj.CIup),
    hide_columns = vars(Adj.CIup),
    pattern = "[{1}, {2}]") %>%
  cols_merge(
    columns=vars(Cochran.Q, Cochran.Q.p.val),
    hide_columns = vars(Cochran.Q.p.val),
    pattern = "{1}, p={2}") %>%
  cols_label(
    Raw.r = "Pearson's r",
    Adj.CIlow = "Adjusted 95% CI",
    p.val= "p-value",
    Adj.p.val="Adjusted p-value",
    Cochran.Q="Cochran's Q") %>% 
  cols_align(
    align = "center")

Study	Outcome	SleepIndicator	N	Pearson's r	p-value	Adjusted p-value	Adjusted 95% CI	Cochran's Q	TOST
Pooled Effect Size	All outcomes	Quality of Acute Sleep	582	0.057	0.102	0.406	[−0.030, 0.143]	24.200, p=0.933	0.135
Pooled Effect Size	All outcomes	Quality of Chronic Sleep	582	0.011	0.745	0.990	[−0.075, 0.097]	24.200, p=0.933	0.088
Pooled Effect Size	All outcomes	Quantity of Acute Sleep	582	0.017	0.622	0.990	[−0.070, 0.103]	24.200, p=0.933	0.095
Pooled Effect Size	All outcomes	Quantity of Chronic Sleep	582	−0.019	0.596	0.990	[−0.105, 0.069]	24.200, p=0.933	0.096

8.2 Sensitivity Analysis 2 : By fitting separate models but with an unstructed variance structure

# Obtaining 4 variance-covariance matrix, one for each sleep indicator

List.Varcovar.QualAcu<-lapply(List.Varcovar.prim_transit1, QualAcuPredictFUNCTION)
List.Varcovar.QualCro<-lapply(List.Varcovar.prim_transit1, QualCroPredictFUNCTION)
List.Varcovar.QuantAcu<-lapply(List.Varcovar.prim_transit1, QuantAcuPredictFUNCTION)
List.Varcovar.QuantCro<-lapply(List.Varcovar.prim_transit1, QuantCroPredictFUNCTION)

V.QualAcu=bldiag(list(
  List.Varcovar.QualAcu[[1]], 
  List.Varcovar.QualAcu[[2]],
  List.Varcovar.QualAcu[[3]],
  List.Varcovar.QualAcu[[4]], 
  List.Varcovar.QualAcu[[5]],
  List.Varcovar.QualAcu[[6]]))
V.QualCro=bldiag(list(
  List.Varcovar.QualCro[[1]], 
  List.Varcovar.QualCro[[2]],
  List.Varcovar.QualCro[[3]],
  List.Varcovar.QualCro[[4]], 
  List.Varcovar.QualCro[[5]],
  List.Varcovar.QualCro[[6]]))
V.QuantAcu=bldiag(list(
  List.Varcovar.QuantAcu[[1]], 
  List.Varcovar.QuantAcu[[2]],
  List.Varcovar.QuantAcu[[3]],
  List.Varcovar.QuantAcu[[4]], 
  List.Varcovar.QuantAcu[[5]],
  List.Varcovar.QuantAcu[[6]]))
V.QuantCro=bldiag(list(
  List.Varcovar.QuantCro[[1]], 
  List.Varcovar.QuantCro[[2]],
  List.Varcovar.QuantCro[[3]],
  List.Varcovar.QuantCro[[4]], 
  List.Varcovar.QuantCro[[5]],
  List.Varcovar.QuantCro[[6]]))

# Perform 4 separate meta-analyses 

Meta.S2.QualAcu<-rma.mv(yi,
                V.QualAcu,  
                random = ~ Outcome | Study, 
                struct="UN", 
                data=ES.Prim[ES.Prim$Predictor=="SleepQualAcu",],
                level=100-(5/4))
Meta.S2.QualCro<-rma.mv(yi,
                V.QualCro,  
                random = ~ Outcome | Study, 
                struct="UN", 
                data=ES.Prim[ES.Prim$Predictor=="SleepQualCro",],
                level=100-(5/4))
Meta.S2.QuantAcu<-rma.mv(yi,
                V.QuantAcu,  
                random = ~ Outcome | Study, 
                struct="UN", 
                data=ES.Prim[ES.Prim$Predictor=="SleepQuantAcu",],
                level=100-(5/4))
Meta.S2.QuantCro<-rma.mv(yi,
                V.QuantCro,  
                random = ~ Outcome | Study, 
                struct="UN", 
                data=ES.Prim[ES.Prim$Predictor=="SleepQuantCro",],
                level=100-(5/4))

#Summary of the meta analyses

List.Meta.S2<-list(Meta.S2.QualAcu, Meta.S2.QualCro, Meta.S2.QuantAcu, Meta.S2.QuantCro)
df.Meta.Res1.S2<-as.data.frame(do.call(rbind, lapply(List.Meta.S2, ExtractMetaFUNCTION)))

df.Meta.S2<- df.Meta.Res1.S2 %>% 
  dplyr::mutate(
    Study="Pooled Effect Size",
    Outcome="All outcomes",
    SleepIndicator=c("Quality of Acute Sleep", "Quality of Chronic Sleep",
                     "Quantity of Acute Sleep","Quantity of Chronic Sleep"),
    Adj.p.val=case_when(
      Adj.p.val>.99~.99,
      Adj.p.val<=.99~Adj.p.val),
    N=sum(N.prim)) %>%
  rowwise() %>%
  dplyr::mutate(TOST=max(TOST1, TOST2)) %>%
  dplyr::select(-c(TOST1,TOST2))

df.Meta.S2<-df.Meta.S2[,c(8:11,1:7,12)]

gt(df.Meta.S2)%>% 
  fmt_number(
    columns = 5:12,
    decimals = 3)%>%
  cols_merge(
    columns=vars(Adj.CIlow, Adj.CIup),
    hide_columns = vars(Adj.CIup),
    pattern = "[{1}, {2}]") %>%
  cols_merge(
    columns=vars(Cochran.Q, Cochran.Q.p.val),
    hide_columns = vars(Cochran.Q.p.val),
    pattern = "{1}, p={2}") %>%
  cols_label(
    Raw.r = "Pearson's r",
    Adj.CIlow = "Adjusted 95% CI",
    p.val= "p-value",
    Adj.p.val="Adjusted p-value",
    Cochran.Q="Cochran's Q") %>% 
  cols_align(
    align = "center")

Study	Outcome	SleepIndicator	N	Pearson's r	p-value	Adjusted p-value	Adjusted 95% CI	Cochran's Q	TOST
Pooled Effect Size	All outcomes	Quality of Acute Sleep	582	0.061	0.090	0.362	[−0.029, 0.150]	3.532, p=0.939	0.141
Pooled Effect Size	All outcomes	Quality of Chronic Sleep	582	0.037	0.350	0.990	[−0.061, 0.133]	10.163, p=0.337	0.124
Pooled Effect Size	All outcomes	Quantity of Acute Sleep	582	0.020	0.578	0.990	[−0.070, 0.110]	4.330, p=0.888	0.101
Pooled Effect Size	All outcomes	Quantity of Chronic Sleep	582	−0.012	0.732	0.990	[−0.102, 0.078]	2.708, p=0.975	0.093

8.3 Sensitivity analysis 3 : One-stage meta-analysis

Meta.S3<-lmerTest::lmer(Moral_TOT ~ SleepType+SleepType:Sleep-1 + (Sleep:SleepType:Outcome-1|Study),
             data=Data_Meta_Raw.long)

## boundary (singular) fit: see ?isSingular

df.Meta.S3<-broom.mixed::tidy(Meta.S3, conf.int = TRUE, conf.level = 1-0.05/4, effects = "fixed")

## Registered S3 method overwritten by 'broom.mixed':
##   method      from 
##   tidy.gamlss broom

8.4 Sensitivity analyses 4 and 5 : Excluding outliers

Function To locate influential points and high leverage using (respectively) cook’s distance and hat values

IdentifyOutlier<-function(x){
  #obtain hat values for each sleep-outcome association
  hatvalues(lm(Moral_tot~SleepQualCro, data=x))->x$HatQualCro
  hatvalues(lm(Moral_tot~SleepQualAcu, data=x))->x$HatQualAcu
  hatvalues(lm(Moral_tot~SleepQuantCro, data=x))->x$HatQuantCro
  hatvalues(lm(Moral_tot~SleepQuantAcu, data=x))->x$HatQuantAcu

  #obtain cooks' distance for each sleep-outcome association
  cooks.distance(lm(Moral_tot~SleepQualCro, data=x))->x$CookQualCro
  cooks.distance(lm(Moral_tot~SleepQualAcu, data=x))->x$CookQualAcu
  cooks.distance(lm(Moral_tot~SleepQuantCro, data=x))->x$CookQuantCro
  cooks.distance(lm(Moral_tot~SleepQuantCro, data=x))->x$CookQuantAcu

  #Identify large hat values (cut off = 3*mean(hat values)
  ifelse(x$HatQualCro>=3*mean(x$HatQualCro), 1, 0)->x$HatQualCroOUT
  ifelse(x$HatQualAcu>=3*mean(x$HatQualAcu), 1, 0)->x$HatQualAcuOUT
  ifelse(x$HatQuantCro>=3*mean(x$HatQuantCro), 1, 0)->x$HatQuantCroOUT
  ifelse(x$HatQuantAcu>=3*mean(x$HatQuantAcu), 1, 0)->x$HatQuantAcuOUT
  
  #Identify large cooks' distance (cut off = 4/sample size)
  ifelse(x$CookQualCro>=4/nrow(x), 1, 0)->x$CookQualCroOUT
  ifelse(x$CookQualAcu>=4/nrow(x), 1, 0)->x$CookQualAcuOUT
  ifelse(x$CookQuantCro>=4/nrow(x), 1, 0)->x$CookQuantCroOUT
  ifelse(x$CookQuantAcu>=4/nrow(x), 1, 0)->x$CookQuantAcuOUT
  x}

Exclusion of participants with a high cook’s distance

Data_Meta_Raw.Slong.split.empty<-split(Data_Meta_Raw.Slong,
                                  list(Data_Meta_Raw.Slong$Study,
                                       Data_Meta_Raw.Slong$Outcome))
 
Data_Meta_Raw.Slong.split<-Filter(function(x) dim(x)[1] > 0,
                                   Data_Meta_Raw.Slong.split.empty)

Data_Meta_S4_transit1<-lapply(Data_Meta_Raw.Slong.split, IdentifyOutlier)
Data_Meta_S4_transit2<-do.call(rbind,Data_Meta_S4_transit1)


#if a participant has a cooks' distance > to the cut off for one sleep-outcome association, he will be considered as an outlier in all associations) 
ID.Cook<-subset(Data_Meta_S4_transit2$ID,
                  Data_Meta_S4_transit2$CookQualAcuOUT==1 |
                  Data_Meta_S4_transit2$CookQualCroOUT==1 |
                  Data_Meta_S4_transit2$CookQuantAcuOUT==1 |
                  Data_Meta_S4_transit2$CookQuantCroOUT==1)

# Exclude influential participants
Data_Meta_Cook<-filter(Data_Meta_Raw.wide, !(ID %in% ID.Cook)) 

# Split the dataset with influential participants excluded
Data_Meta_Cook.split<-split(Data_Meta_Cook, Data_Meta_Cook$Study)

# Obtain sample size for each study
N.Cook<-sapply(Data_Meta_Cook.split, nrow)

# Preparation of the general dataset and obtaining sample size once influential participants  have been excluded
Cor.matrix.Cook<-lapply(Data_Meta_Cook.split, CorFUNCTION)

List.Cor.Varcovar.Cook<-mapply(RmatFUNCTION, Cor.matrix.Cook, N.Cook, SIMPLIFY=FALSE)

# Obtaining effect size for each study, each outcome and each sleep indicator (k=40)
ES.Cook_transit<-do.call(rbind, lapply(List.Cor.Varcovar.Cook, function(x) x$dat))

ES.Cook_transit1<-data.frame(cbind(
  stud.out=row.names(ES.Cook_transit), ES.Cook_transit))

ES.Cook_transit2<-separate(ES.Cook_transit1, stud.out,
                          into=c("Study"))

## Warning: Expected 1 pieces. Additional pieces discarded in 81 rows [1, 2, 3, 4,
## 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...].

ES.Cook<-ES.Cook_transit2 %>%
  rename("Outcome"=var1,
         "Predictor"=var2,
         "Out.Pred"=var1var2) %>% 
  mutate(
    Asso.of.int=case_when(
      str_detect(Outcome, "Sleep") & str_detect(Predictor, "Moral") ~ 1,
      str_detect(Outcome, "Moral") & str_detect(Predictor, "Sleep") ~ 1)) %>%
  filter(Asso.of.int==1)

ES.Cook$N<-rep(N.Cook, c(1*4,1*4,1*4,3*4,2*4,2*4))

#Variance-covariance matrix for each study (N=6)
List.Varcovar.Cook_transit1<-lapply(List.Cor.Varcovar.Cook, function(x) x$V)

List.Varcovar.Cook<-lapply(List.Varcovar.Cook_transit1, FourPredictFUNCTION)

V.Cook=bldiag(list(
  List.Varcovar.Cook[[1]], 
  List.Varcovar.Cook[[2]],
  List.Varcovar.Cook[[3]],
  List.Varcovar.Cook[[4]], 
  List.Varcovar.Cook[[5]],
  List.Varcovar.Cook[[6]]))

# the HCS structure did not achieve convergence. We used a CS structure instead.
Meta.Cook<-rma.mv(yi,V.Cook, 
                  mods=~Predictor-1, 
                  random = ~ Out.Pred | Study, 
                  struct="CS", 
                  data=ES.Cook, level=100-(5/4))

Exclusion of participants with a large hat value

#if a participant has an hat value > to the cut off for one sleep-outcome association, he will be  considered as an outlier in all associations)
ID.Hat<-subset(Data_Meta_S4_transit2$ID,
                  Data_Meta_S4_transit2$HatQualAcuOUT==1 |
                  Data_Meta_S4_transit2$HatQualCroOUT==1 |
                  Data_Meta_S4_transit2$HatQuantAcuOUT==1 |
                  Data_Meta_S4_transit2$HatQuantCroOUT==1)

# Exclude influential participants
Data_Meta_Hat<-filter(Data_Meta_Raw.wide, !(ID %in% ID.Hat)) 

# Split the dataset with influential participants excluded
Data_Meta_Hat.split<-split(Data_Meta_Hat, Data_Meta_Hat$Study)

# Obtain sample size for each study
N.Hat<-sapply(Data_Meta_Hat.split, nrow)

# Preparation of the general dataset and obtaining sample size once influential participants  have been excluded
Cor.matrix.Hat<-lapply(Data_Meta_Hat.split, CorFUNCTION)

List.Cor.Varcovar.Hat<-mapply(RmatFUNCTION, Cor.matrix.Hat, N.Hat, SIMPLIFY=FALSE)



#Obtaining cffect size for each study, each outcome and each sleep indicator (k=40)
ES.Hat_transit<-do.call(rbind, lapply(List.Cor.Varcovar.Hat, function(x) x$dat))

ES.Hat_transit1<-data.frame(cbind(
  stud.out=row.names(ES.Hat_transit), ES.Hat_transit))

ES.Hat_transit2<-separate(ES.Hat_transit1, stud.out,
                          into=c("Study"))

## Warning: Expected 1 pieces. Additional pieces discarded in 81 rows [1, 2, 3, 4,
## 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...].

ES.Hat<-ES.Hat_transit2 %>%
  rename("Outcome"=var1,
         "Predictor"=var2,
         "Out.Pred"=var1var2) %>% 
  mutate(
    Asso.of.int=case_when(
      str_detect(Outcome, "Sleep") & str_detect(Predictor, "Moral") ~ 1,
      str_detect(Outcome, "Moral") & str_detect(Predictor, "Sleep") ~ 1)) %>%
  filter(Asso.of.int==1)

ES.Hat$N<-rep(N.Hat, c(1*4,1*4,1*4,3*4,2*4,2*4))

#Variance-covariance matrix for each study (N=6)
List.Varcovar.Hat_transit1<-lapply(List.Cor.Varcovar.Hat, function(x) x$V)

List.Varcovar.Hat<-lapply(List.Varcovar.Hat_transit1, FourPredictFUNCTION)

V.Hat=bldiag(list(
  List.Varcovar.Hat[[1]], 
  List.Varcovar.Hat[[2]],
  List.Varcovar.Hat[[3]],
  List.Varcovar.Hat[[4]], 
  List.Varcovar.Hat[[5]],
  List.Varcovar.Hat[[6]]))

Meta.Hat<-rma.mv(yi,V.Hat, 
                  mods=~Predictor-1, 
                  random = ~ Out.Pred | Study, 
                  struct="HCS", 
                  data=ES.Hat, level=100-(5/4))

Results of the meta-analysis when influential points are removed from individual studies (Cooks’ distance)

df.Meta.Res1.Cook<-as.data.frame(ExtractMetaFUNCTION(Meta.Cook))

df.Meta.S4<- df.Meta.Res1.Cook %>% 
  dplyr::mutate(
    Study="Pooled Effect Size",
    Outcome="All outcomes",
    SleepIndicator=c("Quality of Acute Sleep", "Quality of Chronic Sleep",
                     "Quantity of Acute Sleep","Quantity of Chronic Sleep"),
    Adj.p.val=case_when(
      Adj.p.val>.99~.99,
      Adj.p.val<=.99~Adj.p.val),
    N=sum(N.Cook)) %>%
  rowwise() %>%
  dplyr::mutate(TOST=max(TOST1, TOST2)) %>%
  dplyr::select(-c(TOST1,TOST2))

df.Meta.S4<-df.Meta.S4[,c(8:11,1:7,12)]

#Summary of the meta analysis
gt(df.Meta.S4)%>% 
  fmt_number(
    columns = 5:12,
    decimals = 3)%>%
  cols_merge(
    columns=vars(Adj.CIlow, Adj.CIup),
    hide_columns = vars(Adj.CIup),
    pattern = "[{1}, {2}]") %>%
  cols_merge(
    columns=vars(Cochran.Q, Cochran.Q.p.val),
    hide_columns = vars(Cochran.Q.p.val),
    pattern = "{1}, p={2}") %>%
  cols_label(
    Raw.r = "Pearson's r",
    Adj.CIlow = "Adjusted 95% CI",
    p.val= "p-value",
    Adj.p.val="Adjusted p-value",
    Cochran.Q="Cochran's Q") %>% 
  cols_align(
    align = "center")

Study	Outcome	SleepIndicator	N	Pearson's r	p-value	Adjusted p-value	Adjusted 95% CI	Cochran's Q	TOST
Pooled Effect Size	All outcomes	Quality of Acute Sleep	508	0.040	0.272	0.990	[−0.051, 0.129]	35.488, p=0.493	0.120
Pooled Effect Size	All outcomes	Quality of Chronic Sleep	508	−0.040	0.262	0.990	[−0.128, 0.049]	35.488, p=0.493	0.119
Pooled Effect Size	All outcomes	Quantity of Acute Sleep	508	0.012	0.733	0.990	[−0.077, 0.101]	35.488, p=0.493	0.092
Pooled Effect Size	All outcomes	Quantity of Chronic Sleep	508	−0.041	0.255	0.990	[−0.130, 0.049]	35.488, p=0.493	0.120

Results of the meta-analysis when participants with an atypical leverage are removed from individual studies (Hat value)

df.Meta.Res1.Hat<-as.data.frame(ExtractMetaFUNCTION(Meta.Hat))

df.Meta.S5<- df.Meta.Res1.Hat %>% 
  dplyr::mutate(
    Study="Pooled Effect Size",
    Outcome="All outcomes",
    SleepIndicator=c("Quality of Acute Sleep", "Quality of Chronic Sleep",
                     "Quantity of Acute Sleep","Quantity of Chronic Sleep"),
    Adj.p.val=case_when(
      Adj.p.val>.99~.99,
      Adj.p.val<=.99~Adj.p.val),
    N=sum(N.Hat)) %>%
  rowwise() %>%
  dplyr::mutate(TOST=max(TOST1, TOST2)) %>%
  dplyr::select(-c(TOST1,TOST2))

df.Meta.S5<-df.Meta.S5[,c(8:11,1:7,12)]

#Summary of the meta analysis
gt(df.Meta.S5)%>% 
  fmt_number(
    columns = 5:12,
    decimals = 3)%>%
  cols_merge(
    columns=vars(Adj.CIlow, Adj.CIup),
    hide_columns = vars(Adj.CIup),
    pattern = "[{1}, {2}]") %>%
  cols_merge(
    columns=vars(Cochran.Q, Cochran.Q.p.val),
    hide_columns = vars(Cochran.Q.p.val),
    pattern = "{1}, p={2}") %>%
  cols_label(
    Raw.r = "Pearson's r",
    Adj.CIlow = "Adjusted 95% CI",
    p.val= "p-value",
    Adj.p.val="Adjusted p-value",
    Cochran.Q="Cochran's Q") %>% 
  cols_align(
    align = "center")

Study	Outcome	SleepIndicator	N	Pearson's r	p-value	Adjusted p-value	Adjusted 95% CI	Cochran's Q	TOST
Pooled Effect Size	All outcomes	Quality of Acute Sleep	531	0.038	0.323	0.990	[−0.059, 0.135]	27.447, p=0.846	0.125
Pooled Effect Size	All outcomes	Quality of Chronic Sleep	531	0.019	0.656	0.990	[−0.086, 0.123]	27.447, p=0.846	0.112
Pooled Effect Size	All outcomes	Quantity of Acute Sleep	531	0.020	0.585	0.990	[−0.073, 0.113]	27.447, p=0.846	0.104
Pooled Effect Size	All outcomes	Quantity of Chronic Sleep	531	0.005	0.904	0.990	[−0.092, 0.102]	27.447, p=0.846	0.092

8.5 Sensitivity analysis 6 : Short durations

#  We removed participants with a task duration lower to a set limit
Data_Meta_Raw.wide.duration<-Data_Meta_Raw.wide %>%
  mutate(
    DurationMinOUT=case_when(
      Duration<=Duration.min~1,
      Duration>Duration.min~0)) %>%
  filter(DurationMinOUT==0)


Data_Meta_S6<-split(Data_Meta_Raw.wide.duration, Data_Meta_Raw.wide.duration$Study)

#Obtaining sample size for each dataset
N.S6<-sapply(Data_Meta_S6, nrow)


#Obtention of Correlation matrix and Variance-Covariance matrix for each study
Cor.matrix.S6<-lapply(Data_Meta_S6, CorFUNCTION)

List.Cor.Varcovar.S6<-mapply(RmatFUNCTION, Cor.matrix.S6, N.S6, SIMPLIFY=FALSE)

#Effect size for each study, each outcome and each sleep indicator (k=40)
ES.S6_transit<-do.call(rbind, lapply(List.Cor.Varcovar.S6, function(x) x$dat))

ES.S6_transit1<-data.frame(cbind(
  stud.out=row.names(ES.S6_transit), ES.S6_transit))

ES.S6_transit2<-separate(ES.S6_transit1, stud.out,
                          into=c("Study"))

## Warning: Expected 1 pieces. Additional pieces discarded in 81 rows [1, 2, 3, 4,
## 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...].

ES.S6<-ES.S6_transit2 %>%
  rename("Outcome"=var1,
         "Predictor"=var2,
         "Out.Pred"=var1var2) %>% 
  mutate(
    Asso.of.int=case_when(
      str_detect(Outcome, "Sleep") & str_detect(Predictor, "Moral") ~ 1,
      str_detect(Outcome, "Moral") & str_detect(Predictor, "Sleep") ~ 1)) %>%
  filter(Asso.of.int==1)

ES.S6$N<-rep(N.S6, c(1*4,1*4,1*4,3*4,2*4,2*4))

#Variance-covariance matrix for each study (N=6)
List.Varcovar.S6_transit1<-lapply(List.Cor.Varcovar.S6, function(x) x$V)

List.Varcovar.S6<-lapply(List.Varcovar.S6_transit1, FourPredictFUNCTION)

V.S6=bldiag(list(
  List.Varcovar.S6[[1]], 
  List.Varcovar.S6[[2]],
  List.Varcovar.S6[[3]],
  List.Varcovar.S6[[4]], 
  List.Varcovar.S6[[5]],
  List.Varcovar.S6[[6]]))

Meta.S6<-rma.mv(yi,V.S6, 
                  mods=~Predictor-1, 
                  random = ~ Out.Pred | Study, 
                  struct="HCS", 
                  data=ES.S6, level=100-(5/4))


#Meta-analyses
df.Meta.Res1.S6<-as.data.frame(ExtractMetaFUNCTION(Meta.S6))

df.Meta.S6<- df.Meta.Res1.S6 %>% 
  dplyr::mutate(
    Study="Pooled Effect Size",
    Outcome="All outcomes",
    SleepIndicator=c("Quality of Acute Sleep", "Quality of Chronic Sleep",
                     "Quantity of Acute Sleep","Quantity of Chronic Sleep"),
    Adj.p.val=case_when(
      Adj.p.val>.99~.99,
      Adj.p.val<=.99~Adj.p.val),
    N=sum(N.S6)) %>%
  rowwise() %>%
  dplyr::mutate(TOST=max(TOST1, TOST2)) %>%
  dplyr::select(-c(TOST1,TOST2))

df.Meta.S6<-df.Meta.S6[,c(8:11,1:7,12)]

#Summary of the meta analysis
gt(df.Meta.S6)%>% 
  fmt_number(
    columns = 5:12,
    decimals = 3)%>%
  cols_merge(
    columns=vars(Adj.CIlow, Adj.CIup),
    hide_columns = vars(Adj.CIup),
    pattern = "[{1}, {2}]") %>%
  cols_merge(
    columns=vars(Cochran.Q, Cochran.Q.p.val),
    hide_columns = vars(Cochran.Q.p.val),
    pattern = "{1}, p={2}") %>%
  cols_label(
    Raw.r = "Pearson's r",
    Adj.CIlow = "Adjusted 95% CI",
    p.val= "p-value",
    Adj.p.val="Adjusted p-value",
    Cochran.Q="Cochran's Q") %>% 
  cols_align(
    align = "center")

Study	Outcome	SleepIndicator	N	Pearson's r	p-value	Adjusted p-value	Adjusted 95% CI	Cochran's Q	TOST
Pooled Effect Size	All outcomes	Quality of Acute Sleep	574	0.093	0.016	0.065	[−0.004, 0.187]	25.237, p=0.910	0.178
Pooled Effect Size	All outcomes	Quality of Chronic Sleep	574	0.056	0.159	0.637	[−0.043, 0.154]	25.237, p=0.910	0.144
Pooled Effect Size	All outcomes	Quantity of Acute Sleep	574	0.032	0.376	0.990	[−0.058, 0.121]	25.237, p=0.910	0.112
Pooled Effect Size	All outcomes	Quantity of Chronic Sleep	574	−0.003	0.934	0.990	[−0.093, 0.087]	25.237, p=0.910	0.084

8.6 Sensitivity analysis 7 : all participants

# we filter the raw datasets to remove only participants with a missing value
Data_Meta_Raw.wide.total<-filter(Data_Meta_Raw_transit.wide, CCA==1) 

Data_Meta_S7<-split(Data_Meta_Raw.wide.total, Data_Meta_Raw.wide.total$Study)


#Obtaining sample size for each dataset
N.S7<-sapply(Data_Meta_S7, nrow)


#Obtention of Correlation matrix and Variance-Covariance matrix for each study
Cor.matrix.S7<-lapply(Data_Meta_S7, CorFUNCTION)

List.Cor.Varcovar.S7<-mapply(RmatFUNCTION, Cor.matrix.S7, N.S7, SIMPLIFY=FALSE)

#Effect size for each study, each outcome and each sleep indicator (k=40)
ES.S7_transit<-do.call(rbind, lapply(List.Cor.Varcovar.S7, function(x) x$dat))

ES.S7_transit1<-data.frame(cbind(
  stud.out=row.names(ES.S7_transit), ES.S7_transit))

ES.S7_transit2<-separate(ES.S7_transit1, stud.out,
                          into=c("Study"))

## Warning: Expected 1 pieces. Additional pieces discarded in 81 rows [1, 2, 3, 4,
## 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...].

ES.S7<-ES.S7_transit2 %>%
  rename("Outcome"=var1,
         "Predictor"=var2,
         "Out.Pred"=var1var2) %>% 
  mutate(
    Asso.of.int=case_when(
      str_detect(Outcome, "Sleep") & str_detect(Predictor, "Moral") ~ 1,
      str_detect(Outcome, "Moral") & str_detect(Predictor, "Sleep") ~ 1)) %>%
  filter(Asso.of.int==1)

ES.S7$N<-rep(N.S7, c(1*4,1*4,1*4,3*4,2*4,2*4))

#Variance-covariance matrix for each study (N=6)
List.Varcovar.S7_transit1<-lapply(List.Cor.Varcovar.S7, function(x) x$V)

List.Varcovar.S7<-lapply(List.Varcovar.S7_transit1, FourPredictFUNCTION)

V.S7=bldiag(list(
  List.Varcovar.S7[[1]], 
  List.Varcovar.S7[[2]],
  List.Varcovar.S7[[3]],
  List.Varcovar.S7[[4]], 
  List.Varcovar.S7[[5]],
  List.Varcovar.S7[[6]]))

Meta.S7<-rma.mv(yi,V.S7, 
                  mods=~Predictor-1, 
                  random = ~ Out.Pred | Study, 
                  struct="HCS", 
                  data=ES.S7, level=100-(5/4))


#Meta-analyses
df.Meta.Res1.S7<-as.data.frame(ExtractMetaFUNCTION(Meta.S7))

df.Meta.S7<- df.Meta.Res1.S7 %>% 
  dplyr::mutate(
    Study="Pooled Effect Size",
    Outcome="All outcomes",
    SleepIndicator=c("Quality of Acute Sleep", "Quality of Chronic Sleep",
                     "Quantity of Acute Sleep","Quantity of Chronic Sleep"),
    Adj.p.val=case_when(
      Adj.p.val>.99~.99,
      Adj.p.val<=.99~Adj.p.val),
    N=sum(N.S7)) %>%
  rowwise() %>%
  dplyr::mutate(TOST=max(TOST1, TOST2)) %>%
  dplyr::select(-c(TOST1,TOST2))

df.Meta.S7<-df.Meta.S7[,c(8:11,1:7,12)]

#Summary of the meta analysis
gt(df.Meta.S7)%>% 
  fmt_number(
    columns = 5:12,
    decimals = 3)%>%
  cols_merge(
    columns=vars(Adj.CIlow, Adj.CIup),
    hide_columns = vars(Adj.CIup),
    pattern = "[{1}, {2}]") %>%
  cols_merge(
    columns=vars(Cochran.Q, Cochran.Q.p.val),
    hide_columns = vars(Cochran.Q.p.val),
    pattern = "{1}, p={2}") %>%
  cols_label(
    Raw.r = "Pearson's r",
    Adj.CIlow = "Adjusted 95% CI",
    p.val= "p-value",
    Adj.p.val="Adjusted p-value",
    Cochran.Q="Cochran's Q") %>% 
  cols_align(
    align = "center")

Study	Outcome	SleepIndicator	N	Pearson's r	p-value	Adjusted p-value	Adjusted 95% CI	Cochran's Q	TOST
Pooled Effect Size	All outcomes	Quality of Acute Sleep	617	0.084	0.019	0.075	[−0.005, 0.172]	36.504, p=0.445	0.163
Pooled Effect Size	All outcomes	Quality of Chronic Sleep	617	0.041	0.285	0.990	[−0.055, 0.136]	36.504, p=0.445	0.126
Pooled Effect Size	All outcomes	Quantity of Acute Sleep	617	0.039	0.274	0.990	[−0.049, 0.126]	36.504, p=0.445	0.117
Pooled Effect Size	All outcomes	Quantity of Chronic Sleep	617	−0.001	0.983	0.990	[−0.090, 0.088]	36.504, p=0.445	0.081