apaTables also supports repeated measures ANOVA tables and repeated measures by between measures (mixed) ANOVA tables via ezANOVA output from the ez package.
Repeated Measures: 2-way design
Prior to begining we ‘open’ the apaTables, tidyverse, and ez packages using the library command as per below.
library(apaTables)
library(tidyverse)
library(ez)
In this example, we used the drink_attitude dataset from Field et al. (2012). As before, this dataset is built into apaTables so it does not need to be loaded using read_csv or read_sav. The dataset (drink_attitude_wide) can be inspected with the glimpse command below (this provides information similar to SPSS variable view). We use the “wide” descriptor in the name of the dataset to remind us that the data is in in the “wide” format where one row contains all the data for one person. There are 20 participants so there are 20 rows.
glimpse(drink_attitude_wide)
## Observations: 20
## Variables: 10
## $ participant <fct> P1, P2, P3, P4, P5, P6, P7, P8, P9, P10, P11, P...
## $ beer_positive <int> 1, 43, 15, 40, 8, 17, 30, 34, 34, 26, 1, 7, 22,...
## $ beer_negative <int> 6, 30, 15, 30, 12, 17, 21, 23, 20, 27, -19, -18...
## $ beer_neutral <int> 5, 8, 12, 19, 8, 15, 21, 28, 26, 27, -10, 6, 4,...
## $ wine_positive <int> 38, 20, 20, 28, 11, 17, 15, 27, 24, 23, 28, 26,...
## $ wine_negative <int> -5, -12, -15, -4, -2, -6, -2, -7, -10, -15, -13...
## $ wine_neutral <int> 4, 4, 6, 0, 6, 6, 16, 7, 12, 14, 13, 19, 14, 21...
## $ water_positive <int> 10, 9, 6, 20, 27, 9, 19, 12, 12, 21, 33, 23, 21...
## $ water_negative <int> -14, -10, -16, -10, 5, -6, -20, -12, -9, -6, -2...
## $ water_neutral <int> -2, -13, 1, 2, -5, -13, 3, 2, 4, 0, 9, 5, 0, 4,...
The dataset represents an ANOVA with two repeated measures factors. It is a 3 drinks (wine, beer, water) X 3 imagery (positive, negative, neutral) design. The dataset uses 10 columns to represent this design. The first column contains a code for each participant (P1, P2, etc.) Note that the participant column is a factor as indicated by the < fctr > descriptor. The participant column must be a factor which can, as noted previously, be created by the as_factor command.
Each of the remaining 9 columns contain atittude ratings for a single drink/imagery combination cell. The naming convention for column names is critical to the workflow we describe below. Each column name represents the level of the drink variable separated using an underscore ( _ ) from a level of the imagery variable. For example the first column is called beer_positive representing the cell with a combination of the the beer level of drink and the positive level of imagery. Each value in this column represents attitude for a participant in this cell.
In order to conduct a repeated measures ANOVA we need to convert the dataset from the wide format (where each row represents a person) to the long format (where each row represents a single observation from a person). We do this with a series of commands below.
First we rearrange the data so that each row represents a single observation from a person using the gather command from the tidyverse package. The command creates a new dataset where there is a participant column, a cell column that indicates the cell an observation came from (e.g., beer_positive) and an attitude column that contains the attitude rating. In this dataset there is one row per observation. Note that in the command below we use, beer_positive:water_neutral to select all the columns in the dataset between/including beer_positive to water_neutral (i.e., all nine columns). The columns must be in a block in the dataset without other variables interspersed between them.
drink_attitude_long <- gather(data = drink_attitude_wide,
key = cell, value = attitude,
beer_positive:water_neutral,
factor_key=TRUE)
In this new long dataset each of the 20 particpants has nine observations (i.e., nine cells) so there are 180 rows. The problem with data resulting from the above command is that the levels of drink and imagery are combined into a single column called cell. That is cell contains the information in the form beer_positive (i.e., the previous column names for each cell). We can split those levels off into their own columns with the names drink and imagery and turn those columns into factors with the commands below:
drink_attitude_long <- separate(data = drink_attitude_long,
col = cell, into = c("drink","imagery"),
sep = "_", remove = TRUE)
drink_attitude_long$drink <- as_factor(drink_attitude_long$drink)
drink_attitude_long$imagery <- as_factor(drink_attitude_long$imagery)
We can inspect the final long version of the dataset with glimpse again:
glimpse(drink_attitude_long)
## Observations: 180
## Variables: 4
## $ participant <fct> P1, P2, P3, P4, P5, P6, P7, P8, P9, P10, P11, P12,...
## $ drink <fct> beer, beer, beer, beer, beer, beer, beer, beer, be...
## $ imagery <fct> positive, positive, positive, positive, positive, ...
## $ attitude <int> 1, 43, 15, 40, 8, 17, 30, 34, 34, 26, 1, 7, 22, 30...
This output reveals the new data structure. There are now only four columns where each row represents a single observation for a participant. As before, the first column is factor column representing participant. But each participant is represented nines times in this column now - because each participant has nine observations (i.e., one in each of the nine cells). Most importantly, we now have a single column indicating level of drink, a single column indicating level of imagery, and a column indicating the attitude rating.
You can see the first few rows in the more familiar format below:
head(drink_attitude_long)
## # A tibble: 6 x 4
## participant drink imagery attitude
## <fct> <fct> <fct> <int>
## 1 P1 beer positive 1
## 2 P2 beer positive 43
## 3 P3 beer positive 15
## 4 P4 beer positive 40
## 5 P5 beer positive 8
## 6 P6 beer positive 17
Prior to conducting the analysis we set the contrasts as per Field et al. (2012).
alcohol_vs_water <- c(1, 1, -2)
beer_vs_wine <- c(-1, 1, 0)
negative_vs_other <- c(1, -2, 1)
positive_vs_neutral <- c(-1, 0, 1)
contrasts(drink_attitude_long$drink) <- cbind(alcohol_vs_water, beer_vs_wine)
contrasts(drink_attitude_long$imagery) <- cbind(negative_vs_other, positive_vs_neutral)
Then we use the ezANOVA command from the ez package to conduct the repeated measures ANOVA. Be sure to include all of the commands below including the options command which ensures that the output has sufficent number of digits (i.e., numbers after the decimal) for apaTables.
options(digits = 10)
drink_attitude_results <- ezANOVA(data = drink_attitude_long,
dv = .(attitude), wid = .(participant),
within = .(drink, imagery),
type = 3, detailed = TRUE)
Now we make the table based on the output:
apa.ezANOVA.table(drink_attitude_results,
table.number = 11,
filename="Table11_APA.doc")
Repeated Measures and Independent Groups: 3-way design
Prior to begining we ‘open’ the apaTables, tidyverse, and ez packages using the library command as per below.
library(apaTables)
library(tidyverse)
library(ez)
In this example, we used the drink_attitude dataset from Field et al. (2012). As before, this dataset is built into apaTables. The dataset (dating_wide) can be inspected with the glimpse command below (this provides information similar to SPSS variable view). We use the “wide” descriptor in the name of the dataset to remind us that the data is in in the “wide” format where one row contains all the data for one person. There are 20 participants so there are 20 rows.
## Observations: 20
## Variables: 11
## $ participant <fct> P01, P02, P03, P04, P05, P06, P07, P08, P09, P...
## $ gender <fct> Male, Male, Male, Male, Male, Male, Male, Male...
## $ attractive_high <int> 86, 91, 89, 89, 80, 80, 89, 100, 90, 89, 89, 8...
## $ average_high <int> 84, 83, 88, 69, 81, 84, 85, 94, 74, 86, 91, 90...
## $ ugly_high <int> 67, 53, 48, 58, 57, 51, 61, 56, 54, 63, 93, 85...
## $ attractive_some <int> 88, 83, 99, 86, 88, 96, 87, 86, 92, 80, 88, 95...
## $ average_some <int> 69, 74, 70, 77, 71, 63, 79, 71, 71, 73, 65, 70...
## $ ugly_some <int> 50, 48, 48, 40, 50, 42, 44, 54, 58, 49, 54, 60...
## $ attractive_none <int> 97, 86, 90, 87, 82, 92, 86, 84, 78, 91, 55, 50...
## $ average_none <int> 48, 50, 45, 47, 50, 48, 50, 54, 38, 48, 48, 44...
## $ ugly_none <int> 47, 46, 48, 53, 45, 43, 45, 47, 45, 39, 52, 45...
The dataset represents an ANOVA with one independent groups factor and two repeated measures factors. It is a 2 gender (male, female) by 3 looks (attrative, average, ugly) X 3 personality (high, some, none) design. The dataset uses 11 columns to represent this design. The first column contains a code for each participant (P1, P2, etc.) Note that the participant column is a factor as indicated by the “< fctr >” descriptor. The second column is used to indicate the gender of each participant. Both the participant and gender columns are factors.
Each of the remaining 9 columns contain atittude ratings for a single looks/personality combination cell. The naming convention for column names is critical to the workflow we describe below. Each column name represents the level of the looks variable separated using an underscore ( _ ) from a level of the personality variable. For example the first column is called attractive_high representing the cell with a combination of the the attractive level of looks and the high level of personality. Each value in this column represents date rating for a participant in this cell.
In order to conduct a mixed between/within ANOVA we need to convert the dataset from the wide format (where each row represents a person) to the long format (where each row represents a single observation from a person) – taking into account the independent groups factors (gender). We do this with a series of commands below.
First we rearrange the data so that each row represents a single observation from a person using the gather command from the tidyverse package. The command creates a new dataset where there is a participant column, a cell column that indicates the cell an observation came from (e.g., attractive_high) and an date_rating column. Note that in the command below we use, attractive_high:ugly_none to select all the columns in the dataset between/including attractive_high to ugly_none (i.e., all nine columns). The columns must be in a block in the dataset without other variables interspersed between them.
dating_long <- gather(data = dating_wide,
key = cell, value = date_rating,
attractive_high:ugly_none,
factor_key = TRUE)
In this new long dataset each of the 20 particpants hasn nine observations (i.e., nine cells) in addition to their gender classification. As a result, there are 180 rows in the long version of this dataset. The problem with data resulting from the above command is that the levels of looks and personality are combined into a single column called cell. That is cell contains the information in the form beer_positive (i.e., the previous column names for each cell). We can split those levels off into their own columns with the names “looks” and “personality” and turn those columns into factors with the commands below:
dating_long <- separate(data = dating_long,
col = cell, into = c("looks","personality"),
sep = "_", remove = TRUE)
dating_long$looks <- as_factor(dating_long$looks)
dating_long$personality <- as_factor(dating_long$personality)
We can inspect the final long version of the dataset with glimpse again:
## Observations: 180
## Variables: 5
## $ participant <fct> P01, P02, P03, P04, P05, P06, P07, P08, P09, P10, ...
## $ gender <fct> Male, Male, Male, Male, Male, Male, Male, Male, Ma...
## $ looks <fct> attractive, attractive, attractive, attractive, at...
## $ personality <fct> high, high, high, high, high, high, high, high, hi...
## $ date_rating <int> 86, 91, 89, 89, 80, 80, 89, 100, 90, 89, 89, 84, 9...
This output reveals the new data structure. There are now only five columns where each row represents a single observation for a participant. As before, the first column is factor column representing participant. But each participant is represented nines times in this column now - because each participant has nine observations (i.e., one in each of the nine cells). The second column contains the gender factor. The third column indicates the level of looks, and the fourth column indicates the level of personality. The fith column indicates the date_rating.
You can see the first few rows in the more familiar format below:
## # A tibble: 6 x 5
## participant gender looks personality date_rating
## <fct> <fct> <fct> <fct> <int>
## 1 P01 Male attractive high 86
## 2 P02 Male attractive high 91
## 3 P03 Male attractive high 89
## 4 P04 Male attractive high 89
## 5 P05 Male attractive high 80
## 6 P06 Male attractive high 80
Prior to conducting the analysis we set the contrasts as per Field et al. (2012).
some_vs_none <- c(1, 1, -2)
hi_vs_av <- c(1, -1, 0)
attractive_vs_ugly <- c(1, 1, -2)
attractive_vs_average <- c(1, -1, 0)
contrasts(dating_long$personality) <- cbind(some_vs_none, hi_vs_av)
contrasts(dating_long$looks) <- cbind(attractive_vs_ugly, attractive_vs_average)
Then we use the ezANOVA command from the ez package to conduct the repeated measures ANOVA. Be sure to include all of the commands below including the options command which ensures that the output has sufficent number of digits (i.e., numbers after the decimal) for apaTables.
options(digits = 10)
dating_results <-ezANOVA(data = dating_long,
dv = .(date_rating), wid = .(participant),
between = .(gender), within = .(looks, personality),
type = 3, detailed = TRUE)
Now we make the table based on the output:
dating_table <- apa.ezANOVA.table(dating_results,
filename = "Table12_APA.doc",
table.number = 12)
print(dating_table)
