Semeon Risom | Eyetracking Pipeline

Data collection

md skfsd sd lmfkfds mdskfsd sdl mfkfds mdskfs d sdlmfkfds mdskfsd sdl mfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sd lmfkfds mdskfsd sdlmfkfds mdskfsd sdl mfkfds mdskfsd sdl mfkfds mdskfsd sdlmfkfds mds kfsd sdlmfkfds mdskfsd sd lmfkfds md skfsd sdlmf kfds mdskfsd sd lmfkfds mdskfsd sdlm fkfds mdskfsd sdlm fkfds md skfsd sdl mfkfds mdskfsd sdl mfkfds mdskfsd sdlmfkfds mdskfsd sd lmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskf sd sdl mfkfds mdskfsd sdlmf kfds mds kfsd sdl mfkfds mdskfsd sdlmfkfds.

Data processing

								
								#run correlation matrix
								df_corr = df[['dp_bias','n_dp_valid','var_dp_bias',
											'gaze_bias','n_gaze_valid','var_gaze_bias','final_gaze_bias',
											'rrs_brooding','cesd_score',
											'm_rt','m_diff_dotloc','m_diff_stim',
											'luminance']].loc[df['nested'] == 'subject']
								
								#file
								file = 'corr_matrix'
								method = 'spearman'
								title = string.capwords('%s correlation coefficient matrix (p-value).'%(method))
								
								path = path_['output'] + "/analysis/html/%s.html"%(file)
								corr_matrix = plot.corr_matrix(config=config, df=df_corr, path=path, title=title, method=method)
								del path, corr_matrix, file, title, method

Code for creating Correlation Matrix

mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds

Scatterplot (x=m_diff_dotloc, y=n_dp_valid)

mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds mdskfsd sdlmfkfds

Model

index	term	B	SE	t	df	Pr(>\|t\|)
1	(Intercept)	6.0775	0.1605	37.8551	139.448161	0.0
2	osmsos	0.6379	0.2513	2.538	136.000295	0.0123
3	trialTypepofa	0.0138	0.0214	0.6454	27049.127397	0.5187
4	TrialNum	0.1126	0.0467	2.4106	136.999935	0.0173

Table: Linear Mixed Model Regression for Stimulus Onset Error (N = 138)

index	sigma	logLik	AIC	BIC	REMLcrit	df.residual
1	1.760331	-54598.524964	109213.049928	109278.774094	109197.049928	27316

Table: Linear Mixed Model Fit by REML (Laplace Approximation) ['lmer']. This table summarizes effects on onset error rate with trial number, operating system, and stimulus.

The assumptions for the model are:

Residuals are normally distributed (Q-Q Plot).
Homogeneity of variance (Residual vs Fitted Plot).
Independent observations

The following post-hoc analysis were run:

Individual Trend Plot.
Group Trend Plot.
Q-Q Plot.
Residual vs Fitted Plot.

Participants with 'Dotloc' or 'Stimulus' Onset Error median above 3SD (n = [17, 25, 49, 54, 59, 77, 80, 89, 112, 123, 138, 140, 150, 153, 180, 182, 185, 212, 221, 248, 250, 256, 262, 269, 292, 294, 298, 319, 999999, 111111, 156], 18.7%) were excluded from analysis (see methods).

We employed linear mixed effects models with random intercepts and slopes using the lmer() function in the lme4 R package (R Core Team, 2013; Bates, Mächler, Bolker, & Walker, 2015). For our model, Operating System, Stimulus (IAPS, POFA), and Trial. were included as fixed effects. Random effects for Trial, and Participant. were included in the model to account for their respective variation in their slopes and intercepts. Stimulus Onset Error was used as the outcome measure.

Individual Trend Plot of the Difference Between Expected and True Onset Time for Stimulus Onset Error (nested by subject:trial, window=5).

Each individual subjects individual trend is indicated here. Participants with 'Dotloc' or 'Stimulus' onset error rate 3 SD above the median are indicated here with a semi-opaque line. The graph has been clipped at y = 200 for displaying purposes.

Trend Plot of the Difference Between Expected and True Onset Time for Stimulus Onset Error (nested by subject:trial).

Data is either unbinned (c,d) or binned into 33 discrete evenly-sized groups (a,b). The model is still fit using the original data. No participants have been excluded for this analysis. The binned graph has been clipped at y = 1000 for displaying purposes.

The Normal Q-Q plot compares the standardized residuals against the theoretical quantiles from a standard normal distribution. If the model residuals are normally distributed, then the points on this graph will be plotted in a generally straight line.

							
								effects = {}
								
								#----load data
								p = path_['output'] + "/analysis/error.csv"
								#df_error = pd.read_csv(p_error, float_precision='high')
								df_ = pd.read_csv(p, float_precision='high')
								
								#----parameters            
								# dependent variable
								y = 'diff_stim','diff_dotloc'
								# fixed effects
								effects['fixed'] = {
									'os': 'categorical',
									'trialType': 'categorical',
									'TrialNum': 'factorial'
								}
								# random effects
								effects['random'] = {
									'TrialNum': 'factorial',
									'participant': 'factorial',
								}
								
								#----save data for access by R and for calculating dwell time
								csv = "onset_data.csv"
								
								#----run model
								# path
								p = path_['output'] + "/analysis/html/model/lmer/"
								# formula
								f = "sqrt(%s) ~ os + trialType + TrialNum + (1+TrialNum|participant)"%(_y)
								# run
								lmer_, lmer_result, lmer_r, html = model.lmer(config=config_, df=df_, y=_y, f=f, 
								exclude=exclude, csv=csv, path=p, effects=effects)
									
								#-----delete
								del y, _y, f, csv, p

Linear Mixed Model Regression for Stimulus Onset Error (N = 138)

Building a pipeline for processing and analyzing eyetracking data

January 21st, 2019

Data collection

Data processing

Model