Procrastination and Participation in Telephone Surveys

Procrastination

Recently published studies on procrastination have demonstrated its relevance not only to psychological research but also with regard to several societal problems (Steel, 2010). Klingsieck (2013: 26) defined procrastination as ‘the voluntary delay of an intended and necessary and/or (personally) important activity, despite expecting potential negative consequences that outweigh the positive consequences of the delay.’ Procrastination, as the tendency to delay intended important activities, is highly variable. Individuals with a strong tendency towards procrastination will tend to procrastinate in many situations, whereas individuals with a weak tendency towards procrastination are rather good at following their originally intended path of activities. Following common explanation of different values of personality traits (Westhoff and Kluck, 2014), individuals having an average tendency towards procrastination, their procrastination behavior should vary from one activity or situation to the other (e.g. individuals will procrastinate in some situations or on some tasks and not on others). Procrastination can appear in many areas of life. Klingsieck (2013) examined the areas of science and work, everyday routines and duties, health, leisure, family and partnership, and social contacts. Procrastination occurred in all six areas of life. It was more typical in academic and professional areas, in everyday processes and obligations, and in the health area, but less frequent in the areas of leisure, family and partnership as well as social contacts. Similarly, in a sample of undergraduate students, academic studies were frequently interrupted in order to sleep, read or watch television (Pychyl et al., 2000). Finally, procrastination in the US was often observed in the preparation of the annual tax declaration (Mennig, 2019).

According to Steel (2007), procrastination can be characterized as a form of self-regulatory failure that may lead to poorer performance and reduced well-being. In his meta-analysis he identified task aversiveness, task delay, self-efficacy, impulsiveness, and conscientiousness with its facets of self-control, distractibility, organization, and achievement motivation as strong and consistent predictors of procrastination. Influences of neuronal characteristics or genetic predispositions on these personal characteristics and personality traits can be assumed as well (Eysenck, 1967; Birbaumer and Schmidt, 2010; Heckhausen and Heckhausen, 2018). Distractibility, for example, is characterized by being easily distracted and having one’s attention diverted easily (Kasper et al., 2015). Individuals with a higher level of distractibility might drop whatever they are doing more likely to engage in other activities. Distracting events can often be characterized by a change in flow of thoughts which is preceded by an emotionally arousing cue (e.g. orienting reflex, Sokolov, 1963; Heckhausen and Heckhausen, 2018; Klinger, 1996, 1999). Extreme forms of distractibility can be found in clinical samples, e.g. children with autism spectrum disorder (ASD) or with attention-deficit/hyperactivity disorder (ADHD). Both disorders show strong genetic and environmental risk factors (Rommelse et al., 2011; Taurines et al., 2012). When important activities are deferred in favor of less important ones, and the activities actually performed do not correspond to the individual’s intention to achieve the more important goal, this postponement is deemed to be procrastination. Thus, the term procrastination is negatively connoted. Some definitions emphasize that the deferred activities are perceived as aversive by the procrastinator. The activities executed instead are often short-term and can be done quickly (Rist et al., 2006). Participation in a telephone survey could be such a ‘fast-to-do’ alternative.

The tendency to procrastinate certain activities is regarded as a relatively stable, transversal behavioral disposition. In population samples from the United States, Australia, and the United Kingdom to whom standardized procrastination questionnaires were administered, approximately 20% of the respondents could be described as chronic procrastinators (Ferrari et al., 2005). According to other studies, 40% of the participants had experienced financial loss (Ferrari et al., 1995) or other adverse effects (Aitken, 1982) during the past year because of procrastination.

Research on procrastination is frequently focused on academic procrastination among college and university students, as procrastinatory behavior is very common among this population (Yockey, 2016; Pychyl et al., 2000). In the sample of Day et al. (2000: 120) 32% of the individuals were severe procrastinators, Ellis and Knaus (1977) reported about surveys in which even 95% to 99% of the student respondents identified themselves as procrastinators. This is remarkable because procrastination has been found to be related to poor academic performance (Yockey, 2016; Klingsieck, 2013; Ferrari and Emmons, 1995), lower self-efficacy (Steel, 2007), and higher stress levels (Tice and Baumeister, 1997).

According to Klingsieck (2013), who examined procrastinatory behaviors in six different life domains in an online sample of 260 students, procrastination is not a general disposition but rather a domain-specific behavioral tendency. However, confirmation of this finding in non-student samples is still lacking.

Procrastination was recently assessed in a large sample of the general German population in a face-to-face survey (Beutel et al., 2016). The prevalence of procrastination was found to be highest in the 14-29-year-old cohort. Only in this cohort were men found to procrastinate more than women. Furthermore, procrastination was associated with reduced life satisfaction and other negatively connoted characteristics.

The relation between survey participation and procrastination was first addressed by Gabler et al. (2012), who investigated the association between the tendency to procrastinate and whether individuals aborted participation in a telephone survey. The authors found that individuals who completed the interview were much less inclined to procrastinate than those who dropped out. However, for survey research, and for the explanation of nonresponse, many more questions arise in the context of procrastination. If procrastination is a more or less stable behavioral disposition, it is obvious that it will play an important role in the decision-making situation with regard to the costs and benefits of participating in a survey. Thus, when deciding whether or not to participate, participation could be regarded as a welcome alternative to the primary activity and thus be deemed beneficial. This would certainly be in line with the recent conceptualization of procrastination as a ‘short-term mood repair’ strategy (Sirois and Pychyl, 2013). According to this notion, individuals with a tendency to procrastinate will turn mostly to activities that have more hedonistic value than the intended primary activity. This may well apply to survey participation: for example, individuals may decide to participate in a survey because they are interested in the survey topic or because they welcome the social interaction as a means of procrastination (see also Meier et al., 2016).

However, regarding survey participation, postponing activities can be beneficial as well. This is possible in two ways: first, a postponed activity is carried out at a later point in time with a higher degree of care. In regard to survey participation this could be the case if a participant is preoccupied with other activities at the moment of the call, and decides to make an appointment to participate in the survey at a later point in time. Survey participation can then take place with a higher level of concentration. This is not to be called procrastination, but strategic delay. Second, participation in a survey can also be seen as a welcome distraction (procrastination). This could be beneficial, if a strenuous activity is interrupted and can be continued afterwards with new vigor.

Research Questions

To address these issues, we first examined whether and how procrastination could be assessed in a quick and easy way in survey research. Therefore, our study presents data from a large and high-quality telephone sample of the German general population to whom a short scale for measuring procrastination was administered.

Second, we examined how procrastination was related to four different activities – namely, working, relaxing, satisfying basic needs, and doing housework – and whether these activities were interrupted in order to participate in the telephone interview.

All (potential) respondents are engaged in some kind of activity when they receive the survey interviewer’s call. This may be a planned, primary activity, such as working or studying, or an activity of a more procrastinatory nature, such as using Facebook (Meier et al., 2016; Hinsch and Sheldon, 2013; Reinecke et al., 2016). If the call is unscheduled and the individual spontaneously participates in the survey, survey participation itself can be understood as procrastination. However, it is also possible that the respondent rates participation as an activity with a higher benefit. By contrast, if participation in the survey has been arranged by appointment, it is not procrastination but rather a thoroughly planned primary activity.

Therefore, we expected that procrastination would play a different role with regard to survey participation, depending on the activity that was interrupted by the call from the survey firm. Sociodemographic characteristics of the respondents, survey mode (landline vs. mobile phone), and survey time were expected to be further influencing factors.

Participants and Procedure

The results presented here are based on a sample survey of the general German population, CELLA 2, ¹ in which 3,007 participants were interviewed. For our analysis we had 2,943 valid cases (age 16–93 years, M = 43.41; 48% female). Participants were interviewed about their telephone usage behavior. The questionnaire also included several items aimed at measuring data quality (e.g., question-order effects, social desirability, and response stability; see Häder, 2012; Häder et al., 2009). Of the 2,943 interviews, 1,471 were conducted via landline phone and 1,472 via mobile phone. The landline sample was drawn from the universe of possible landline numbers in Germany using simple random sampling (Gabler and Häder, 2016). The sampling frame comprised 139,366,300 numbers, from which 31,358 numbers were selected. A modified random digit dialing (RDD) method was used to select potential participants for the mobile phone survey (Gabler et al., 2012: 154ff.). The sampling frame comprised 197,490,000 mobile phone numbers, from which 44,330 numbers were selected. Both samples were drawn by GESIS – Leibniz Institute for the Social Sciences in Mannheim, Germany. The same instrument was used nationwide for both samples. The fieldwork was carried out in summer 2010 by a commercial survey research firm and lasted six weeks. The mean duration of the interviews was 12.33 minutes. The following response rates (RR) according to AAPOR standards were realized: RR3_l = 0.148 for the landline phone sample and RR3_m= 0.117 for the mobile phone sample (Schneiderat and Schlinzig, 2012: 124). However, despite these low response rates, the quality of the collected data proved satisfactory. Schneiderat and Schlinzig (2012: 131) concluded: ‘In summary, it can be stated that despite low response rates, comparisons of the CELLA 2 data to official reference statistics showed that our samples performed in representing the characteristics of individual subgroups of the survey’s target population. The integration of a mobile sample by applying a dual frame approach nearly always leads to better sample quality.’

Questionnaire

In the CELLA 2 questionnaire (see Häder and Häder, 2009), the short form of the Procrastination Scale for Students (PFS-4) was included. This scale measures procrastination as a behavioral tendency that is regarded as widely stable across different situations (see Glöckner-Rist et al., 2014). The authors of the scale developed and tested it in the framework of an online survey with university student samples. For the short form, they selected from a 12-item version of the scale four items that showed both high discriminatory and high convergent validity. The short form was developed to enable the economical measurement of procrastination in multi-topic surveys, as well as for screening purposes. Klingsieck (2013) used this scale successfully in an online survey of university students. However, CELLA 2 was a telephone survey directed at the German general population. In contrast to online surveys with university student samples, the items and the response options were communicated verbally to the respondents (see Table 1). It was therefore necessary to first examine the quality of the instrument in this new setting and with this new sample. Furthermore, it had to be determined whether the instrument could also be used in a survey on both landline and mobile phones. There are many differences in landline and mobile phone surveys which can influence the possibility to conduct a specific survey and influence the quality of the data collected (Häder and Kühne, 2009: 170f). For example, the location at which the respondents are reached in (at home versus anywhere in the world), other individuals being present at the time of the interview (known individuals at home versus strangers), distractions by the respective environment, technical problems (with mobile phone surveys the network coverage, the battery performance, the transmission quality, etc.) and experience in handling technical equipment (in mobile phone surveys this is especially relevant for older people, see Brust et al., 2016). Another aspect is accessibility: Landline telephone surveys, and also face-to-face surveys only take place at times, when the respondents are at home. Surveys on mobile phones, on the other hand, can be conducted at all times. For this reason alone, the accessibility is different for different survey modes.

OPEN IN VIEWER

Table 1.

Descriptive statistics of the items of the German-language version of the PFS-4 and results of the principal components analysis

Item	Mean	Standard Deviation	FactorLoadings	n
Ich schiebe den Beginn von Aufgaben bis zum letzten Moment hinaus. [I put off beginning tasks until the last minute.]	2.11	1.125	0,843	2936
Ich warte mit dem Beginn einer Arbeit so lange, dass es mir schwer fällt, sie noch rechtzeitig zu beenden. [I put off beginning a task so long that I find it hard to complete it on time.]	2.34	1.100	0.828	2938
Beim Bearbeiten einer Aufgabe merke ich, dass ich sie schon viel früher hätte erledigen können. [When I’m working on a task, I realize that I could have accomplished it much sooner.]	2.55	1.095	0,744	2930
Ich schaffe es erst auf den letzten Drücker meine Aufgaben zu erledigen. [I only manage to complete my tasks at the last minute.]	2.12	1.161	0,808	2932

Note: n = sample size; unweighted data; our functionally equivalent translation of the German-language items in square brackets; The following instruction was used in the telephone interviews: ‘Now let’s come to difficulties that you may have when dealing with everyday things. When you think back over the last two weeks, how often did you behave in the following way: almost never, rarely, sometimes, frequently, or almost always?’ (Min. = 1, Max. = 5).

To avoid order effects, the items were randomly rotated in CELLA 2. Descriptive statistics of the German-language version of the PFS-4 are shown in Table 1.

Statistical Analyses to Test the Dimensionality and Reliability of the PFS-4

First, we conducted exploratory factor analysis to assess the dimensionality of the scale. This analysis revealed a one-factor solution with the factor loadings shown in Table 1. The factor analyses calculated separately for each mode (landline and mobile phone) also yielded one-factor solutions, as well as very similar findings with respect to the individual factor loadings (results not shown). The subsequent reliability test revealed a satisfactory value of 0.820 for Cronbach’s alpha. Further improvement of the value by deleting items was not possible. Cronbach’s alpha was 0.822 for the landline phone sample and 0.818 for the mobile phone sample.

Thus, an additive weighted sum index could be calculated from the four items. The index was right-skewed, with a mode of 0.81. It was smaller than the median (1.78) and the mean (M = 1.83; Min = 1; Max = 5). Such a right-skewed distribution was also expected by the authors of the scale in the case of its use in a general population survey (see Glöckner-Rist, 2014).

To further assess the quality of the scale, we conducted item response theory (IRT) analysis using the R eRm package (Mair et al., 2016) and Mplus (Version 7; Muthén and Muthén, 2012).

IRT models offer advantages compared to classical approaches to decide whether the data meet assumptions. Therefore, they are useful to determine the quality of a measure. To assess the dimensionality and reliability of the PFS-4, we fitted one-parameter logistic (1PL) and two-parameter logistic (2PL) models to the data. If data fit the more parsimonious 1PL model, it can be assumed that the measure in question is characterized by unidimensionality, monotonicity, local stochastic independence, generalizability across samples, generalizability across items, sufficient statistics, and specific objectivity of comparisons. By contrast, 2PL models have fewer preconditions. However, if data fit only a 2PL model, information can be obtained only about unidimensionality, monotonicity and local stochastic independence (Moosbrugger, 2006; Fox and Jones, 1998). The partial credit model (PCM; Masters, 1982) for polytomous data as a 1PL model and the graded response model (GRM; Samejima, 1969) for polytomous data as a 2PL model were calculated.

The requirements for a 1PL model fit are quite strict and are usually tested by verifying whether the assumption of generalizability across samples holds. Therefore, using Andersen’s likelihood ratio test (Andersen, 1973), it is tested statistically whether the estimated parameters of the model are identical across different samples. According to common guidelines, the model can be expected to fit the data if at least four comparisons across subsamples are non-significant (p >.01, resulting in a .05 level of significance considering alpha error cumulation).

The requirements for a 2PL model fit are less strict. However, it is nonetheless difficult to decide how useful a 2PL model is to explain the data. In our analysis, we adopted an approach described by Glöckner-Rist (2014) in which model fit is estimated using a generalized structural equation model (SEM) implemented in Mplus (Version 7, Muthén and Muthén, 2012). First, a 2PL probit IRT graded response model (GRM; Samejima, 1969) is estimated and its general model fit is evaluated; second, subgroup comparison (i.e., multigroup analysis) is used to evaluate measurement invariance across subgroups as an indication of model fit.

Model estimation and general model fit: Besides using a chi-square test to assess model fit, we also used typical SEM fit indices. In chi-square testing, a non-significant result is regarded as an indicator of model fit. However, chi-square test results are influenced by sample size (Tucker and Lewis, 1973). Therefore, it is reasonable to use also goodness-of-fit indices that are considered to be relatively robust to sample size differences. To evaluate model fit, the comparative fit index (CFI), the Tucker-Lewis index (TLI), the root mean square error of approximation (RMSEA), and the weighted root mean square residual (WRMR) were used. According to common guidelines for the assessment of model fit (see Marsh et al., 1988; Hu and Bentler, 1999; Marsh et al., 2004; Yu, 2002; Yu and Muthén, 2002), CFI and TLI values greater than .90 and .95 indicate acceptable and excellent fit, respectively; RMSEA values less than .05 and .08 indicate close and reasonable fit, respectively; and WRMR values less than .90 and 1.00 indicate close and reasonable fit, respectively. Following Glöckner-Rist (2014), we also used the weighted least squares means and variance adjusted estimator (WLSMV) with theta parameterization, which often produces more stable and less inflated parameter estimates than the ordinary WLS estimator (see Flora and Curran, 2004: 473; Glöckner-Rist, 2014).

Multigroup analyses: To test for subsample differences and to evaluate measurement invariance across subgroups as an indication of model fit, multiple-group analyses were conducted as follows: The most restrictive model (factor loadings and item thresholds constrained to be equal across two groups) was compared with two less restrictive models, where either only the factor loadings or only the item thresholds were constrained to be equal (Glöckner-Rist, 2014; Lang et al., 2011; see also Marsh et al., 2013). These three different models of measurement invariance were compared within four different groups: low versus high score (median split); male versus female; young adults versus older adults (median split); low versus high education. In order to evaluate differences between the estimated models and to identify the most parsimonious model that best fits the data, Bentler (1990) suggested testing nested models using chi-square difference testing. When using the WLSMV estimator, chi-square difference testing is not possible in the regular way. Therefore, an adjusted procedure implemented in Mplus must be used (Muthén and Muthén, 2012). However, chi-square difference testing is also dependent on sample size (Brannick, 1995). Therefore, the examination of changes in fit indices is used as an alternative to this procedure, as well (Cheung and Rensvold, 1999; Chen, 2007). According to Chen (2007), a more parsimonious model is supported if the CFI change is smaller than .01 or the RMSEA change is smaller than .015. According to Marsh et al. (2009), unchanged, or even improved, TLI and RMSEA values compared to the less restrictive model are a conservative criterion for the most parsimonious model.

Statistical Analysis to Test Associations between Procrastination and other Survey Variables

We used binary logistic regression to test associations between procrastination and other survey variables. To this end, several variables were transformed into dummy variables. Calculations were performed using SPSS 25.0 (IBM, 2017).

IRT Analysis of PFS-4

Using the 1PL approach, the model did not fit the data. The comparison of the estimated parameters (Cond. LL = -7107.956) of Masters’ partial credit model (PCM; Masters, 1982) across four subgroups revealed significant differences in three cases (Andersen LR-Test: low score vs. high score: LR/chi-square df = 146.53/15, p = .001; male vs. female: LR/chi-square df = 22.378/15, p = .098, n.s.; low age vs. high age: LR/chi-square df = 31.993/15, p = .006; low education vs. high education: LR/chi-square df = 134.386/15, p = .001). Therefore, the item parameters of the four items of the PFS-4 were not identical across different samples. There were differences between respondents with low and high education, between lower- and higher-age respondents, and between respondents with a low and with a high score on the measure. This third difference implied that the PFS-4 functions differently in respondents who exhibit few procrastination behaviors compared to those with a tendency to procrastinate.

Using the 2PL approach, the model fit was more appropriate. Samejima’s graded response model (GRM; Samejima, 1969) showed excellent fit in the total sample (chi-square/df = 1.913/2; p = .384, n.s.; n = 2907; CFI/TLI = 1.000/1.000; RMSEA = .001; WRMR = .174). The comparison of the estimated parameters of Samejima’s GRM (Samejima, 1969) across the four subgroups revealed the following: whereas almost all CFI, TLI, and RMSEA indices showed excellent fit, and the comparison of CFI, TLI, and RMSEA between the most restrictive model and the two less restrictive models revealed almost no differences across all comparisons, chi-square difference testing and the evaluation of the WRMR did reveal differences (see Table 2). Similar to the 1PL approach, differences were found between respondents with low and high education and between respondents with a low and with a high score on the measure. This also implies that items function differently in respondents who exhibit different levels of procrastinatory behavior.

OPEN IN VIEWER

Table 2.

Comparison of fit of the IRT PL2 model (GRM) multiple group analysis across four different groups (median split): low vs. high score, male vs. female, young adults vs. older adults, low vs. high education

Model		WLSMV/df	p fit p diff	CFI	TLI	RMSEA	WRMR
(1) Score: FL free – TH constant		57/15	.001	.960	.968	.044	1.488
(2) Score: FL constant – TH free		8/4	.075	.996	.987	.028	.583
	(3) vs. (1)	15/4	.006	.008	.002	.001	.255
	(3) vs. (2)	61/15	.001	.044	.017	.015	1.160
(3) Score: FL constant – TH constant		70/19	.001	.952	.970	.043	1.743
(4) Sex: FL free – TH constant		18/15	.258	1.000	1.000	.012	.656
(5) Sex: FL constant – TH free		5/4	.278	1.000	1.000	.014	.284
	(6) vs. (4)	9/4	.058	.001	.001	.008	.403
	(6) vs. (5)	24/15	.062	.001	.001	.006	.775
(6) Sex: FL constant – TH constant		30/19	.058	.999	.999	.020	1.059
(7) Age: FL free – TH constant		16/15	.393	1.000	1.000	.006	.614
(8) Age: FL constant – TH free		3/4	.508	1.000	1.000	.001	.229
	(9) vs. (7)	4/4	.355	.001	.001	.002	.196
	(9) vs. (8)	16/15	.400	.001	.001	.004	.581
(9) Age: FL constant – TH constant		19/19	.426	1.000	1.000	.004	.810
(10) Edu: FL free – TH constant		70/15	.001	.995	.996	.050	1.326
(11) Edu: FL constant – TH free		9/4	.071	1.000	.999	.028	.371
	(12) vs. (10)	13/4	.012	.001	.001	.009	.341
	(12) vs. (11)	56/15	.001	.004	.002	.013	1.296
(12) Edu: FL constant – TH constant		65/19	.001	.996	.997	.041	1.667

Note: n = 2,907; unweighted data, WLSMV/df = weighted least squares means and variance adjusted estimator/degrees of freedom; p_fit = chi-square test to evaluate model fit; p_diff = chi- square difference test between two models; CFI = comparative fit index; TLI = Tucker-Lewis index; RMSEA = root mean square error of approximation; WRMR = weighted root mean square residual; FL = factor loadings; TH = thresholds.

In sum, IRT analysis revealed a measure that shows a sufficient fit with regard to the graded response IRT model (GRM). Therefore, the PFS-4 should be suitable for use in survey research. However, our results also show that further efforts to improve the measure are needed. In particular, the fact that items functioned differently in respondents with different levels of procrastination could be seen as serious problem with regard to the correct interpretation of results.

Relationship between Procrastination and Survey Participation

To examine the relationship between procrastination and survey participation, we first distinguished between different activities that preceded survey participation; second, we differentiated between ‘cold’ contacts leading to the interview and interviews arranged by appointment. A pre-arranged appointment should reduce the perceived cost of participation and lead to a higher willingness to interrupt a particular current activity. With regard to the activities that were spontaneously interrupted in order to participate in the interview, compared to the activities that were being engaged in when a scheduled call (by appointment) requesting an interview was received, the results were as follows:

Of the 2,943 interviews conducted, 2,363 were spontaneous and 580 were conducted on the basis of a pre-arranged appointment. At the beginning of the interview, the target individuals were asked about the activity they had been performing immediately before the call. The four activities shown in Table 3 were those that were most frequently interrupted. ²

OPEN IN VIEWER

Table 3.

Activities interrupted for the telephone interview (column percentages)

	Spontaneous(‘cold’)	By Appointment
Relaxing or engaging in a leisure activity	46.2	39.1
Working – incl. studying	16.8	21.9
Housework, e.g. cleaning, washing dishes, etc.	16.1	18.3
Basic needs, e.g. eating, drinking, sleeping	13.7	14.8

Note: unweighted data.

It appears that spontaneous telephone requests for an interview actually have a chance of immediate success, especially if the target individual is just relaxing.

Target individuals are engaged in different activities at the time of the call. The nature of the activity, and the possible consequences of interrupting it, should influence participation. The activities may be both primary activities and activities performed at that moment due to a tendency to procrastinate. In the context of a complex cost-benefit analysis of whether or not to participate in a telephone survey, and for the willingness to interrupt a particular activity for a telephone survey, many aspects, such as the survey mode (via landline or mobile phone), spontaneous versus scheduled participation, the time of day, the demographic characteristics of the target individual, and the tendency to procrastinate, may play a role. Logistic regression models were estimated for the four most frequently interrupted activities (Table 4). In these models, the characteristics of the interviewed individual, the characteristics of the interview situation, and whether the call was spontaneous or scheduled were used as predictors.

OPEN IN VIEWER

Table 4.

Results of the binary logistic regression – Exp (B) – the explanation of different activities interrupted for the interview (0 = no, 1 = yes)

	Interrupted activity
Variables	Working	Relaxing	Basic Needs	Housework
Constant	0.933	1.124	0.030***	0.066***
Years of education	1.070*	0.951**	1.113**	0.976
Sex (1 = male; 2 = female)	0.419***	0.947	0.989	2.270***
Age (in years)	0.991**	1.004	1.002	1.005
Procrastination index	0.789**	0.984	1.387***	0.975
Spontaneous (0 = no; 1 = yes)	0.531***	1.304**	1.204	0.885
Mode (1 = landline; 2 = mobile)	1.582***	0.918	0.960	0.957
Before 6 p.m.	0.974	0.811**	0.712**	1.552*
After 8 p.m.	1.078	1.433**	0.644	0.454*
Nagelkerkes R2	0.009	0.017	0.025	0.069
n	2.081	2.081	2.081	2.081

Note: *p ≤ 0.10 **p ≤ 0.05, ***p ≤ 0.01; general-regression (GREG) weighted data (see Gabler et al., 2012); SPSS 25, procedure: binary logistic regression, method: enter.

Procrastination was a significant predictor for survey participation in the two models ‘Working’ and ‘Basic Needs’.

Education, sex, age, survey mode, spontaneous versus scheduled participation, and the time at which the call was received were also important predictors in the different models. The following results were obtained:

Individuals who showed a higher willingness to interrupt work or study in order to participate in the survey were more educated, more frequently male, and younger. In the decision whether to participate or not, the tendency to procrastinate played a negative, that is, cost-enhancing role. Individuals who tended to procrastinate were less likely to be reached by telephone during working hours. On the other hand, a previously given commitment to participate in the survey appears to have resulted in a cost reduction. Those individuals who had agreed in advance to participate in the survey had also been surveyed more frequently via mobile phone. This could mean that the requirements of the primary activity ‘working’ were not very demanding and it was therefore interrupted in favor of the new primary activity ‘survey participation’. This could have been the case, for example, for business travelers, who usually have only a mobile phone available. The time of the call, on the other hand, did not matter.

Individuals who had been relaxing immediately before the interview began and who agreed to interrupt this activity had more often an appointment for the interview, and these interviews more often took place after 8 p.m. These individuals also reported fewer years of formal education. The survey mode (landline or mobile) had no influence.

Individuals who were satisfying basic needs (e.g., eating, drinking, or sleeping) when they received the call, and who agreed to interrupt these activities, were mainly more highly educated individuals with a tendency to procrastinate. In addition, these individuals were reached mainly after 6 p.m., whereas the interview commenced more often before 6 p.m., if participants had interrupted their housework in order to participate in the survey. Those participants were more often female.

Finally, participants were asked whether they had been distracted during the survey interview. Individuals with a tendency to procrastinate were distracted more often during the interview (t =3.63 , df =2.908, distracted M = 1.91, n = 810; not distracted M = 1.80, n = 2.100, p = .000).