ezANOVA error message

[Kitti Ban]

Hi, 

I'm trying to run a 2 (between-subjects) x 3 (within-subjects) x 3 (within-subjects) ANOVA, however, I keep getting the following messages: 

Warning: Converting "ID" to factor for ANOVA.

Warning: You have removed one or more levels from variable "offered". Refactoring for ANOVA.

Warning: Data is unbalanced (unequal N per group). Make sure you specified a well-considered value for the type argument to ezANOVA().

Error in ezANOVA_main(data = data, dv = dv, wid = wid, within = within,  : 

  One or more cells is missing data. Try using ezDesign() to check your data.

> 

I used the following command: 

proposer_anova <- ezANOVA(data = proposer_data_frame, 

                   dv = .(percentage_offered), 

                   wid = .(ID), 

                   between = .(native),

                   within = .(trustworthiness, offered),

                   type = 2, 

                   detailed = TRUE, 

                   return_aov = TRUE)

Thank you for any help with this issue, I'm really stuck because of this. 

All the best,

Kitti 

This is my data file (each column seems to be pushed to the left):

ID	native	offered	trustworthiness	percentage_offered	mean_reaction_time

1	1	TRUE	ambiguous	low	13.64	2480.000
2	1	TRUE	ambiguous	middle	8.16	1809.000
3	1	TRUE	fair	middle	8.16	1671.250
4	1	TRUE	unfair	high	100.00	2765.000
5	1	TRUE	unfair	low	86.36	1673.158
6	1	TRUE	unfair	middle	83.67	3181.220
7	2	TRUE	ambiguous	low	22.22	6877.500
8	2	TRUE	ambiguous	middle	7.84	6451.250
9	2	TRUE	fair	high	100.00	2044.583
10	2	TRUE	fair	low	77.78	6601.429
11	2	TRUE	fair	middle	92.16	2736.766
12	3	TRUE	ambiguous	high	33.33	3880.000
13	3	TRUE	ambiguous	low	44.00	2713.091
14	3	TRUE	ambiguous	middle	78.95	3080.833
15	3	TRUE	fair	high	66.67	3333.000
16	3	TRUE	fair	low	16.67	1830.000
17	3	TRUE	fair	middle	7.89	4561.667
18	3	TRUE	unfair	low	52.00	2886.538
19	3	TRUE	unfair	middle	13.16	2544.800
20	4	FALSE	fair	high	100.00	1026.000
21	4	FALSE	fair	middle	100.00	1499.831
22	5	FALSE	ambiguous	high	50.00	2536.111
23	5	FALSE	ambiguous	low	50.00	4336.500
24	5	FALSE	ambiguous	middle	48.00	2470.292
25	5	FALSE	fair	high	38.89	5534.429
26	5	FALSE	fair	low	33.33	2602.000
27	5	FALSE	fair	middle	42.00	2835.810
28	5	FALSE	unfair	high	11.11	3501.500
29	5	FALSE	unfair	low	50.00	14762.000
30	5	FALSE	unfair	middle	10.00	8654.800
31	6	FALSE	ambiguous	low	65.00	2518.923
32	6	FALSE	ambiguous	middle	65.38	1820.794
33	6	FALSE	fair	low	35.00	1925.571
34	6	FALSE	fair	middle	32.69	1723.706
35	6	FALSE	unfair	middle	5.00	1225.000
36	7	TRUE	fair	middle	100.00	1841.194
37	8	FALSE	ambiguous	low	28.26	2735.462
38	8	FALSE	ambiguous	middle	50.00	2728.308
39	8	FALSE	fair	low	19.57	2500.889
40	8	FALSE	fair	middle	23.08	2181.333
41	8	FALSE	unfair	low	52.17	2283.500
42	8	FALSE	unfair	middle	26.92	1856.857
43	9	TRUE	ambiguous	middle	7.69	6107.500
44	9	TRUE	fair	high	100.00	3801.450
45	9	TRUE	fair	middle	92.31	3537.917
46	10	TRUE	ambiguous	high	12.50	1224.000
47	10	TRUE	ambiguous	low	12.50	1926.000
48	10	TRUE	ambiguous	middle	15.62	2480.700
49	10	TRUE	fair	high	63.16	3705.500
50	10	TRUE	fair	low	78.57	1822.455
51	10	TRUE	fair	middle	64.06	2123.024
52	10	TRUE	unfair	high	31.58	1403.500
53	10	TRUE	unfair	low	14.29	2635.500
54	10	TRUE	unfair	middle	20.31	2584.154
55	11	FALSE	ambiguous	high	35.71	3313.000
56	11	FALSE	ambiguous	low	50.00	1625.333
57	11	FALSE	ambiguous	middle	55.77	2853.793
58	11	FALSE	fair	high	42.86	1936.833
59	11	FALSE	fair	low	20.00	2824.000
60	11	FALSE	fair	middle	17.31	2190.667
61	11	FALSE	unfair	high	21.43	2444.667
62	11	FALSE	unfair	low	33.33	1684.000
63	11	FALSE	unfair	middle	26.92	2245.214
64	12	TRUE	fair	high	100.00	1821.000
65	12	TRUE	fair	middle	100.00	1833.843
66	13	TRUE	fair	high	100.00	1427.263
67	13	TRUE	fair	middle	100.00	1189.265
68	14	FALSE	ambiguous	high	11.11	5896.000
69	14	FALSE	ambiguous	low	23.53	5764.750
70	14	FALSE	ambiguous	middle	36.96	5615.294
71	14	FALSE	fair	high	88.89	4037.750
72	14	FALSE	fair	low	23.53	4724.000
73	14	FALSE	fair	middle	45.65	6287.095
74	14	FALSE	unfair	low	52.94	5821.333
75	14	FALSE	unfair	middle	17.39	4549.250
76	15	FALSE	ambiguous	high	12.00	3225.000
77	15	FALSE	ambiguous	low	50.00	5252.000
78	15	FALSE	ambiguous	middle	14.29	7657.500
79	15	FALSE	fair	high	88.00	2435.591
80	15	FALSE	fair	middle	78.57	3376.182
81	15	FALSE	unfair	low	80.00	5826.750
82	15	FALSE	unfair	middle	7.14	5223.000
83	16	FALSE	ambiguous	high	25.00	2839.000
84	16	FALSE	ambiguous	low	100.00	1801.000
85	16	FALSE	ambiguous	middle	33.33	3898.150
86	16	FALSE	fair	high	50.00	3965.333
87	16	FALSE	fair	low	25.00	5146.000
88	16	FALSE	fair	middle	36.67	4196.682
89	16	FALSE	unfair	high	33.33	3649.500
90	16	FALSE	unfair	low	66.67	3424.500
91	16	FALSE	unfair	middle	30.00	3738.389
92	17	FALSE	ambiguous	high	100.00	3310.000
93	17	FALSE	ambiguous	low	61.54	1347.375
94	17	FALSE	ambiguous	middle	58.14	1817.000
95	17	FALSE	fair	high	100.00	1677.000
96	17	FALSE	fair	low	15.38	2290.750
97	17	FALSE	fair	middle	25.58	2057.182
98	17	FALSE	unfair	high	33.33	1780.000
99	17	FALSE	unfair	low	23.08	2096.167
100	17	FALSE	unfair	middle	16.28	3682.000
101	18	TRUE	fair	high	100.00	1110.143
102	18	TRUE	fair	low	100.00	1908.000
103	18	TRUE	fair	middle	100.00	1608.053
104	19	TRUE	ambiguous	high	100.00	4173.000
105	19	TRUE	ambiguous	low	16.67	8632.000
106	19	TRUE	ambiguous	middle	7.14	3792.667
107	19	TRUE	fair	high	91.67	2141.136
108	19	TRUE	fair	low	83.33	1894.600
109	19	TRUE	fair	middle	92.86	2181.385
110	19	TRUE	unfair	high	11.11	4211.000
111	20	FALSE	fair	high	100.00	1546.083
112	20	FALSE	fair	low	100.00	1537.600
113	20	FALSE	fair	middle	100.00	2037.436
114	21	FALSE	ambiguous	high	14.29	2134.000
115	21	FALSE	ambiguous	low	100.00	2503.000
116	21	FALSE	ambiguous	middle	21.67	2470.923
117	21	FALSE	fair	high	92.31	1500.083
118	21	FALSE	fair	middle	76.67	1890.457
119	21	FALSE	unfair	middle	11.11	2100.000
120	22	TRUE	ambiguous	high	31.58	3251.500
121	22	TRUE	ambiguous	low	35.00	13025.286
122	22	TRUE	ambiguous	middle	48.48	2529.688
123	22	TRUE	fair	high	57.89	2099.000
124	22	TRUE	fair	middle	24.24	2385.500
125	22	TRUE	unfair	high	10.53	2490.500
126	22	TRUE	unfair	low	65.00	4453.923
127	22	TRUE	unfair	middle	27.27	2053.444
128	23	TRUE	ambiguous	low	100.00	2728.000
129	23	TRUE	ambiguous	middle	30.43	3448.857
130	23	TRUE	fair	high	100.00	2114.000
131	23	TRUE	fair	middle	68.12	3374.660
132	23	TRUE	unfair	middle	3.57	9638.000
133	24	TRUE	fair	high	100.00	1359.619
134	24	TRUE	fair	middle	100.00	1381.433
135	25	TRUE	ambiguous	high	33.33	5585.000
136	25	TRUE	ambiguous	low	59.26	4978.500
137	25	TRUE	ambiguous	middle	18.42	4096.143
138	25	TRUE	fair	high	85.71	3759.500
139	25	TRUE	fair	low	40.74	3209.364
140	25	TRUE	fair	middle	81.58	4489.710
141	26	FALSE	ambiguous	high	34.48	5026.600
142	26	FALSE	ambiguous	low	30.77	4018.750
143	26	FALSE	ambiguous	middle	40.00	4121.417
144	26	FALSE	fair	high	44.83	4283.231

Showing 1 to 16 of 473 entries

[Mike Lawrence]

Run the following and send a copy of the resulting plot:

ezDesign(
    data = proposer_data_frame,
    y = ID,
    x = native,
    row = trustworthiness,
    col = offered
 )

[Kitti Ban]

Hi Mike, 

Thank you for your reply, I'm sorry for the delay, I had some other problems with R I needed to sort first to be able to run the code. I attached the resulting graph. 

Thanks again, 

Kitti

[Mike Lawrence]

Ok, it looks like you have missing data. If you run this:

ezDesign(
    data = proposer_data_frame, 
    y = ID, 

    x = offered,
    row = trustworthiness,
 )

You'll notice that some rows don't have blank (grey) areas. Those are condition combinations where a participant doesn't have any data.

--
Mike Lawrence
Graduate Student
Department of Psychology & Neuroscience
Dalhousie University

~ Certainty is (possibly) folly ~

--
You received this message because you are subscribed to the Google Groups "ez4r" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ez4r+uns...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Kitti Ban]

Thank you for your fast reply! Do you know any way I could run the ANOVA despite the missing combinations?

[Mike Lawrence]

No, it is not possible to use ANOVA when data are missing. See here for discussion of alternatives: https://groups.google.com/d/msg/ez4r/PbtPgKtH_zI/XWZhhXQWAwAJ

--
Mike Lawrence
Graduate Student
Department of Psychology & Neuroscience
Dalhousie University

~ Certainty is (possibly) folly ~

[Kitti Ban]

Thank you for the suggestion and all of the help!  

[Kitti Ban]

I started following the instructions given in that post and wrote the following: 

# between-subjects variable

contrasts(proposer_data_frame$native) = contr.sum

# within subject bariables

contrasts(proposer_data_frame$offered) = contr.helmert

contrasts(proposer_data_frame$trustworthiness) = contr.helmert

# ezMixed

fit = ezMixed(data = proposer_data_frame[!is.na(proposer_data_frame$percentage_offered),],

  dv=.(percentage_offered), 

  random=.(ID), 

  fixed=.(offered, native, trustworthiness))

  

# Post-hoc

preds = ezPredict(fit$models$`native:offered:trustworthiness`$unrestricted)

For the last line, I got the following error message: 

Error in ezPredict(fit$models$`native:offered`$unrestricted) : 

  ezPredict does not know how to handle fits of class "lmerModLmerTest"

Could you please help me with this analysis? 

Thank you once again,

Kitti

On Wednesday, 25 July 2018 23:22:44 UTC+1, Mike Lawrence wrote:

[Mike Lawrence]

Is it possible that you loaded the package "lmerTest"? It overrides things that ez uses behind the scenes. Try running this code to load an updated version of ezPredict and let me know if it works:

ezPredict <-
function(
    fit
    , to_predict = NULL
    , numeric_res = 0
    , boot = TRUE
    , iterations = 1e3
    , zero_intercept_variance = FALSE
){
    fit_class = class(fit)[1]
    if((fit_class=='mer')|(fit_class=='glmerMod')|(fit_class=='lmerMod')|(fit_class=='lmerModLmerTest')){
        data = attr(fit,'frame')
        vars = as.character(attr(data,'terms'))
        dv = vars[2]
        vars = gsub('\\(.+?\\) ?\\+','',vars[3])
        vars = gsub('\\+ ?\\(.+?\\)','',vars)
        # vars = gsub('\\(.+\\)','',vars[3])
        vars = unlist(strsplit(vars,'+',fixed=T))
        vars = str_replace_all(vars,' ','')
        vars = vars[nchar(vars)>0]
        these_terms = vars
        vars = vars[!str_detect(vars,':')]
        vars = unlist(strsplit(vars,'*',fixed=T))
        # vars = as.character(attr(attr(data,'terms'),'variables'))
        # dv = as.character(vars[2])
        # vars = vars[3:length(vars)]
    }else{
        if(fit_class%in%c('gam','bam')){
            data = fit$model
            randoms = NULL
            for(i in fit$smooth){
                if(class(i)[1]=='random.effect'){
                    randoms = c(randoms,i$term)
                }
            }
            vars = as.character(attr(attr(data,'terms'),'variables'))
            dv = as.character(vars[2])
            vars = vars[3:length(vars)]
            vars = vars[!(vars%in%randoms)]
            BY = vars[str_detect(vars,'BY')]
            vars = vars[!str_detect(vars,'BY')]
        }else{
            stop(paste('ezPredict does not know how to handle fits of class "',fit_class,'"',sep=''))
        }
    }
    if(is.null(to_predict)){
        if(length(grep('poly(',vars,fixed=TRUE))>0){
            stop('Cannot auto-create "to_predict" when the fitted model contains poly(). Please provide a data frame to the "to_return" argument.')
        }
        data_vars = vars[grep('I(',vars,fixed=T,invert=T)]
        temp = list()
        for(i in 1:length(data_vars)){
            this_fixed_data = data[,names(data)==data_vars[i]]
            if(is.numeric(this_fixed_data)&(numeric_res>0)){
                temp[[i]] = seq(
                    min(this_fixed_data)
                    , max(this_fixed_data)
                    , length.out=numeric_res
                )
            }else{
                temp[[i]] = sort(unique(this_fixed_data))
                if(!is.numeric(this_fixed_data)){
                    contrasts(temp[[i]]) = contrasts(this_fixed_data)
                }
            }
        }
        to_return = data.frame(expand.grid(temp))
        names(to_return) = data_vars
    }else{
        to_return = to_predict
    }
    data_vars = names(to_return)
    if(fit_class%in%c('gam','bam')){
        for(i in randoms){
            to_return$EZTEMP = data[1,names(data)==i]
            names(to_return)[ncol(to_return)] = i
        }
        for(i in BY){
            to_return$EZTEMP = ''
            for(j in str_split(i,'BY')[[1]]){
                to_return$EZTEMP = paste(to_return$EZTEMP,as.character(to_return[,names(to_return)==j]),sep='')
            }
            to_return$EZTEMP = ordered(to_return$EZTEMP)
            names(to_return)[ncol(to_return)] = i
        }
    }
    to_return$ezDV = 0
    names(to_return)[ncol(to_return)] = dv
    if((fit_class=='mer')|(fit_class=='glmerMod')|(fit_class=='lmerMod')){
        requested_terms = terms(eval(parse(text=paste(
            dv
            , '~'
            , paste(
                these_terms#attr(attr(data,'terms'),'term.labels')
                , collapse = '+'
            )
        ))))
        mm = model.matrix(requested_terms,to_return)
        f = lme4::fixef(fit)
        v = vcov(fit)
        if(zero_intercept_variance){
            v[1,] = 0
            v[,1] = 0
        }
    }else{
        mm <- predict(fit,to_return,type="lpmatrix") # get a coefficient matrix
        for(i in randoms){
            mm[,grep(paste('s(',i,')',sep=''),dimnames(mm)[[2]],fixed=T)] = 0 #zero the subject entry
        }
        f = coef(fit)
        for(i in randoms){
            f[grep(paste('s(',i,')',sep=''),names(f),fixed=T)] = 0 #zero the subject entry
        }
        v = vcov(fit)
        if(zero_intercept_variance){
            v[1,] = 0
            v[,1] = 0
        }
        for(i in randoms){
            row = grep(paste('s(',i,')',sep=''),dimnames(v)[[1]],fixed=T)
            col = grep(paste('s(',i,')',sep=''),dimnames(v)[[2]],fixed=T)
            v[row,] = 0
            v[,col] = 0
        }
    }
    value = mm %*% f
    to_return$value = as.numeric(value[,1])
    tc = Matrix::tcrossprod(v,mm)
    to_return$var = Matrix::diag(mm %*% tc)
    to_return = to_return[,names(to_return) %in% c(data_vars,'value','var')]
    if(boot){
        samples = mvrnorm(iterations,f,v)
        mat = matrix(NA,nrow=nrow(to_return),ncol=iterations)
        for(i in 1:iterations){
            mat[,i] <- mm%*%samples[i,]
        }
        boots = as.data.frame(to_return[,names(to_return) %in% data_vars])
        names(boots) = data_vars
        boots = cbind(boots,as.data.frame(mat))
        boots = melt(
            data = boots
            , id.vars = names(boots)[1:(ncol(boots)-iterations)]
            , variable.name = 'iteration'
        )
        to_return = list(
            cells = to_return
            , boots = boots
        )
    }
    return(to_return)
}

[Kitti Ban]

Indeed, I had the lmerTest package loaded. However, when detaching, I received the error message "Error in str_replace_all(vars, " ", "") : 

  could not find function "str_replace_all"". The function you sent went through without any warnings or errors, however, I'm not entirely sure what it does.  Am I supposed to be looking out for a table similar to the summary table of an ANOVA?