Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation

Hacienda Pública Española / Review of Public Economics, 230-(3/2019): 95-124 © 2019, Instituto de Estudios Fiscales https://doi.org/10.7866/hpe-rpe.19.3.4 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation* DAVID PATIÑO** FRANCISCO GÓMEZ-GARCÍA*** Universidad de Sevilla Received: November, 2016 Accepted: November, 2018 Abstract Traditional neoclassical thought fails to explain questions such as problems of self-control. Behavioural economics have explained these matters on the basis of the intertemporal preferences of individuals and, specifically, the so-called ( β, δ ) model which emphasises present bias. This opens the way to the analysis of new situations in which people can adopt incorrect indecisions that make it necessary for the government to intervene. The literature which has developed the ( β, δ ) model and its implications has generated a categorisation of people that is widely used but which lacks a systematic empirical evaluation. It is important to value the need for this public action. In this article, we develop a method which makes it possible to verify the main implications that this model has to explain the procrastination of university students. Using an experimental time discount task with real monetary incentives, we estimate the students’ β and δ parameters and we analyse their correlation with their answers to a series of questions concerning how they plan to study for an exam. The results are ambiguous given that they back some of the model’s conclusions but reject others, including a number of the most basic ones, such as the relation between present biases and some of the categories of people, these being essential to predict their behaviour. Keywords: Behavioural economics, problems of self-control, welfare analysis, experimental economics. JEL Classification: I210, D90 1. Introduction Economics has been broadening its subjects of analysis to study questions which were previously outside its traditional area of interest. The extension has enabled the introduction * We thank Pablo Brañas-Garza and Antonio M. Espín, the two anonymous referees, and the executive editors for their helpful discussions and comments. ** ORCID ID: 0000-0002-3313-561X. *** ORCID ID: 0000-0002-6430-0331. 96 david patiño and francisco gómez-garcía of a different perspective to that used in other disciplines but has highlighted the difficulties of explaining a series of phenomena through the traditional economic perspective. This has allowed for the reconsideration of numerous questions of traditional economic analysis and has even opened the door to new proposals of economic policies. Behavioural economics has been founded on the study of phenomena that conventional neoclassic economics does not have an explanation for, or whose explanations are unsatisfactory. Its analysis is based on the cognitive limitations of people that do not permit them to assimilate all the information necessary to adopt complex decisions. This often leads them to follow simple rules which disregard a good part of the salient information. For example, they give more relevance to events which take place close to them in space or in time than what they objectively have. Furthermore, it uses behavioural biases found by psychology to understand behavioural economics. Madrian (2014) underscores the importance of this question, as it allows for the locating of market failures in addition to those traditionally considered. But it also opens the way to proposing new formulas of economic policies or carrying out a different valuation of those which are usually applied. Congdon et al. (2011) define three broad categories of psychological biases that can be the source of market failures: imperfect optimisation, limited self-control and non-traditional preferences. Limited selfcontrol, a phenomenon this work is centred on, is manifested in the discrepancy between people’s intentions and their actual behaviour. People frequently plan to behave in a specific way but end up doing so another way. They likely procrastinate, or they modify their choices according to their emotional state, or small barriers, which objectively are not so, are significant impediments of their actions. To disregard the effects of this issues can lead to choosing mistaken instruments of economic policies. For instance, Campbell et al. (2011) note that the effectiveness of the supply of obligatory information as a way of resolving market failures, such as the existence of externalities, is limited if the consumers do not understand the information, if they believe that it is not relevant to the adopting of their decisions, or if they do not know how to access it or use it. Self-control problems can be understood as the incapacity of some subjects to dominate their desires to achieve their aims. Among them, procrastination stands out. This has concerned economists since at least Strotz (1956). According to Akerlof (1991), procrastination takes place when the current costs are unduly stressed in comparison with those of the future. This leads people to postpone tasks without realising that, when it is again the time to do the task, they will put it off again. The most widespread explanation of this way of behaving is based on people’s intertemporal preferences and explains how behaviour is planned in time, and why such a plan is reneged when it implies carrying out tasks that are costly in terms of effort. The decision of the present reduces the future well-being and people later regret their choices. The phenomenon is analogous to an externality towards oneself and it is sometimes denoted as internality. The traditional conclusion of neoclassical economics is that people are the best guarantors of their own interests and supposes that they are the ones who best know how to choose what will improve their well-being. Yet this is not ensured when biases exist. To measure the internalities requires identification of the impact of agents’ choices under their own experienced utility. This is similar to how a traditional externality requires identification of an agent’s impact on the utilities experienced by others. 97 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation The analysis of procrastination has been used to explain phenomena such as drug addiction and, in general, the adoption of numerous habits considered harmful or unhealthy (Read and Van Leeuwen, 1998). In the area of economics, its consequences have been especially studied for decisions with respect to savings (Thaler and Shefrin, 1981). Procrastination has also been used to illustrate why the availability to pay with a credit card grows as postponing the payment reduces the current value of the debt (Prelec and Simester, 2001); explain the functioning of bureaucracies in an alternative way to the agent-principal model (Akerlof, 1991); and show that spectators’ choices of their type of films give rise to biases towards commercial films (Read et al., 1999), to cite only a few outstanding examples. Faced with the design of public actions, the detailed knowledge of these aspects is important as the difference between people’s intentions and their actions can vary as a response to very small changes in the context of their choice (Madrian and Shea, 2001). But furthermore, the degree of self-control depends on the current state of the deciders and their emotions. Elements such as stress, an overloading of information or fear can set off impatience and motivate radical changes in behaviour. According to behavioural economics, in the cases in which many people show cognitive biases or a lack of self-control, the role of the government should not be limited to a minimum, given that people cannot free themselves from the mistakes of their decisions. It is indispensable to know the mechanism which produces these discrepancies to discern when a nudge is necessary (Thaler and Sunstein, 2008). Chetty (2015) points out that the decision to include behavioural elements in economic models must be considered as being more a pragmatic than a philosophical choice. Nevertheless, given the multitude of biases which distance people from the behaviour predicted by conventional models, it is necessary to determine which are decisive and introduce them. To identify the optimal policy requires evaluating the extent to which the utilities experienced by people differ from the decisions that they really adopt. Yet this opens the door to arbitrariness. This is why it is imperative to empirically measure the degree to which utility and decisions are detached from each other, which explains the methods that we propose to use in this work. Specifically, the literature has suggested measuring experienced utility using data on self-reported happiness. This is an analogous approach to that employed in the contingent evaluation methods which assess externalities (Diamond and Hausman 1994). Likewise, the idea has arisen in other articles of calibrating the structural parameters of a model that includes behavioural biases. We employ this notion in the central part of this work. In this line, this article proposes methods to measure people’s degree of error in their decisions and when they do so, as well as to analyse their consequences in terms of wellbeing. Its aim is to analyse a specific reality – the daily activity of university students in preparing a subject – and measure the degree to which this process fits what the theory predicts. To do so, an empirical methodology is introduced which enables this verification to be carried out. The so-called quasi-hyperbolic discounting allows for modelling of the behaviour of people who postpone their decisions or procrastinate. Its use has been generalised and has 98 david patiño and francisco gómez-garcía fostered the development of a typology of people with different behaviours regarding selfcontrol problems. However, the relation between quasi-hyperbolic discounting and different time preferences has not been the object of a systematic empirical evaluation. The main aim of this article consists in verifying whether the model explains the procrastination of a sample of university students when performing their academic activities. To do so, we have carried out two surveys which permit us to find out their study habits and the students’ characteristics. One of them includes a habitual discount task that has enabled to us to infer the students’ time preferences and to characterise them according to the degree of consistency that they present. This information allows for verification of most of the implications of the explanation of self-control problems based on quasi-hyperbolic preferences. Specifically, the relation between present biases, the type of time preferences people have, and their behaviour and the costs of their self-control problems in terms of well-being and poor academic performances are verified. The article’s main conclusion permits the establishment of an inverse empirical relation between the size of the present bias and maintaining behaviours consistent with the students’ time preferences, as the model that we aim to verify predicts. Nonetheless, we have not found a relation with the rest of the categories of people or with the rest of the theory’s implications. We believe that our results are important in that they shed light on the almost non-existent empirical basis of the ( β, δ ) model and its conclusions. On the other hand, our evaluation also gives keys to valuing the extent to which it is necessary to design new action instruments in the educational area. The article is structured in 6 sections, including this introduction. In the second, we review the economic literature that analyses self-control problems, their implications for public policies and their relationship with time preferences. The third indicates the empirical and experimental methodology followed to analyse the questions proposed. The fourth describes how the database was built and carries out a brief analysis of its descriptive statistics. The fifth shows the models which have been used to empirically verify the aspects analysed and extensively analyses the results obtained. The article ends with a conclusions section. 2. The problem of self-control and intertemporal preferences The mainstream economic analysis that assumes that rational people adopt results consistent with their preferences has great difficulties in explaining self-destructive behaviours, for instance drug addiction or compulsive food consumption. Dissatisfaction with the approach, in spite of its attempts1 to explain such phenomena, has fostered the search for alternatives based mainly on concepts common in psychology and framed in the area of behavioural economics. These explanations of lack of self-control have revolved around time preferences and possible shortsighted calculations of the benefits and costs of actions2. Ifcher and Zarghamee (2011) indicate that the psychological framework for lack self-control overlaps with the economic concept of time preference. 99 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation The analysis of the problems of self-control lies within the study of pathological divergences between the choices of people and their preferences. The most accepted explanations are based on the proposal that there are two types of thought: one which gives swift, automatised and unconscious answers; the other is slow thought that is logical and is done consciously (Kanheman, 2001). For example, Bernheim and Rangel (2004) use this framework to analyse drug addiction. For these authors, the mechanisms of semi-automatic answers are beneficial, especially in stable environments, because they generate quick answers in multiple circumstances. Notwithstanding, they can lead to systematic mistakes that can be serious. In their model, people can make decisions “coldly”, imposing cognitive control. This type of decisions results in the choice of the alternative preferred. But there also exists a “hot” mode in which decisions and preferences can differ. Thaler and Shefrin (1981) contemplate a double personal plan in the adoption of decisions to explain self-control problems. Each person has a farsighted-planner and shortsighted-doer nature which maintains a kind of agent-principal relation with divergent interests. The planner obtains utility uniquely through the actions that the executor carries out. The model predicts that people will establish restrictions of their own behaviour mainly in the actions whose benefits and costs are produced at different moments. The actions of the planners can consist of modifying the preferences of the executor, acting on their incentives or limiting their set of possibilities of choice. Gul and Pesendorfer (2001) show, in a similar framework, how temptations can be combatted by establishing limitations to the set of choices. Likewise, Fudenberg and Levin (2006) indicate that this view is compatible with much evidence of magnetic resonance images, as many decision problems can be explained as a game between a sequence of impulsive short-term selves and patient long-term selves. Models based on an agent-principal problem centre their explanation on time preference biases. O’Donoghue and Rabin (1999) explain how people procrastinate. The “long-term self” establishes the plan, but what is commonly called losses of self-control, caused by present biased time preferences, arises. Their effect is that immediate gratifications are valued to a greater extent than if the actions had been carried out at a later moment. This same idea has fostered later versions that have modelled a broad range of decisions, such as saving and drug consumption. This has propagated the need to rethink the explanation of how decisions in time are adopted, which has been dominated by the theory of discounted utility. This theory was developed by Samuelson (1937), who extended Irving Fisher’s previous idea to multiple periods. Discounted utility reduces all motives which lead people to value the future in relation with the present to a unique parameter known as the discount rate. The discount factor enables people to interchange the future utility with that of the present. Koopman (1960) later demonstrated that the model could be obtained from a series of plausible axioms and this model gained in relevance. More formally, the standard model of temporal preference designed by Samuelson (1937) is based on the existence of an exponential temporal discounting rate which is constant over time. For all t, the utility of an individual would be: 100 david patiño and francisco gómez-garcía (1) Where δ € (0, 1] is the discounting factor. If individuals have a bias towards immediacy, it is necessary to weigh the remoteness or nearness of the event. This can be introduced by employing a quasi-hyperbolic temporal discounting model; see Strotz (1956), Phelps and Pollak (1968) and Laibson (1997)3 Concretely, the bias of the preferences is inserted via a function designed by Phelps and Pollak (1968) in the context of intergenerational altruism. This function adds an additional factor to Samuelson’s intertemporal preferences which weigh the utilities obtained in periods following that which is taken as a reference. In this way the model introduces the present bias by overdiscounting the utility obtained in periods subsequent to the reference. We can rewrite the utility function to include such biases as: (2) Where 0 ≤ β , δ ≤ 1, β measures the present bias. If it is close to 1 it hardly exists, that is to say, the now is not especially valued with respect to the afterwards. On the contrary, a β close to 0 indicates an impatience or excessive eagerness to achieve an immediate reward. The model explains the decision to undertake actions whose benefits and costs are generated at different moments. The problem of self-control arises when the discount rate rises at the time of performing the action, generating a recalculation of the total balance of benefits and costs stemming from it. The result may be different to that provided by the longterm discounting rate and cause a change of decision. People do not change their preferences, or at least they do not change them permanently or stably. Once the moment has passed, they return to a stable or reflexive situation. To evaluate the cost of the lack of selfcontrol, the reference is the decisions that a person with time consistent preferences would adopt and which are those that would be chosen in the long run, given their time preference. O’Donoghue and Rabin (1999) propose a classification of people according to their time preferences. People with a present bias have time preferences consistent over time and do not suffer from self-control problems. We can distinguish two types among those who have a present bias. On the one hand, sophisticated people are aware of their bias and of the selfcontrol problems that this will cause them. To avoid them they adopt measures which, in general, consist in carrying out the action before. The result is suboptimal but better than no action4. Naïve people do not foresee that they will suffer self-control problems. They have present biases the same as sophisticated people but, unlike them, they plan the future ignoring their present biases. As they do not adopt any kind of cautionary measure, it is likely that they will support the totality of the costs of well-being. These are due to not adjusting to the planned behaviour which, a posteriori, they would have liked to carry out. O’Donoghue and Rabin (2008) later introduced the category of the partially sophisticated to define people who are aware of their present bias but underestimate its degree. The 101 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation condition can be introduced using a parameter that we can denote by ^ β , which measures the agents’ estimations of the size of their own biases. In the case of a person with consistent time preferences, ^ β = β =1. If a bias exists, β < 1. Naïve people believe that their behaviour will be consistent with their preferences, but actually they have a present bias, therefore, ^ β =1> β . If the agents are sophisticated, they correctly predict their present bias, and therefore their self-control problems, so ^ β = β <1 will occur. Finally, partially sophisticated agents will have ^ β < β <1 as they are aware of their self-control problems but underestimate their magnitude5. The empirical literature has concentrated on testing the relationship between discounting rates and behaviours which reveal a lack of self-control to attempt to underline the lack of coherence of the traditional vision in explaining compulsive behaviours. The most common practice has been to exploit the evidence provided by laboratory or field experiments, which are grounded on some method of inference of people’s temporal preferences. These experiments usually consist of asking the individuals to choose between sums of money, real or fictitious, which are smaller in a close moment in time and greater later6, in order to calibrate when the utility of both is balanced7. For example, Meier and Sprenger (2012) have studied the relationship between present bias and the financial solvency of individuals. Reynolds (2006) explains drug consumption and gambling, Kirby et al. (1999) heroin addiction, Bickel et al. (1999) smoking and Weller et al. (2008) obesity. Reuben et al. (2015) stand out for having a direct relation with our study aim. Their work analyses the relation between time preferences and procrastination through a series of laboratory experiments and field work with a population of students. They estimate the parameters which define the time preferences via a set of tasks of the type indicated in the previous paragraph and the level of trust, cognitive skills and gender are among the controls used. Burks et al. (2012) compare the goodness of different methods of inferring time preferences, contrasting the extent to which the discount factors estimated by each one explain different phenomenon. Specifically, they analyse the accumulation of human capital (Eckel et al., 2007), savings (Ashraf et al., 2006) and academic results (e.g., Shoda et al., 1990). The different estimations use experiments carried out on middle-aged workers with low skill levels. The functional form which best predicts the decisions analysed is quasi-hyperbolic discounting, calculated from a set of choices over sums of money at different moments in time. Another outstanding work is Nardotto’s (2011), which identifies the different categories of people described above, along with their characteristics. It uses a sample of frequencies of access and tariffs contracted by users of a university gymnasium along with their academic qualifications. They build a suboptimal index or “cost” for not fulfilling their own plan that is explained by the people’s characteristics. This divides the people into consistent (or rational), naïve or sophisticated, comparing the planned behaviour with what is finally carried out. To build this fixed classification, a threshold of 25% of mistakes includes the unpredictable motives that prevent fulfilling the plan. This work finds that 40.6% of the people 102 david patiño and francisco gómez-garcía predict well and are catalogued as rational. 51.1% have optimistic preferences with respect to their forecasts of attendance and are catalogued as naïve, and 4.3% are catalogued as sophisticated. Lastly, Wong (2008) analyses the preparation for the final exam of a subject by a group of degree students. He identifies consistent, sophisticated, ingenuous and partially ingenuous individuals using two questionnaires. The first questionnaire is about the amount of study that students consider ideal and which they estimate that they will actually do and is asked halfway through the term. The second questionnaire is done the day of the exam and infers, a posteriori, the amount of study really done. This work finds a small percentage of consistent students who trust in fulfilling their ideal study plan and indeed do so. Among the inconsistent, three behaviour patterns are identified. The naïve who predict the fulfilment of their ideal study plan but do not fulfil it. The sophisticated who fulfil their forecasts but whose plans do not correspond with what is ideal. The author interprets this behaviour with an awareness of future self-control problems and the design of a plan to minimise their consequences. Lastly, the partially naïve are aware that their self-control problems will lead them to not fulfilling their ideal plans and so they also design plans which try to compensate for this, but they do not fulfil them. Wong (2008) employs the delay foreseen in the ideal plan to measure the degree of temporal inconsistency and the delay foreseen in the chosen plan as a measure of the individual’s degree of sophistication. The article concludes that any delay, foreseen or not, has negative effects on the academic results, even controlling for the time really dedicated to studying, which registers those caused by reasons other than self-control problems. It underlines that sophisticated individuals do not manage to reduce the negative effects of self-control problems. The article interprets this result as being a consequence of these individuals’ poor distribution of study time. Our work follows Wong’s closely, but additionally introduces an eliciting of time preferences. In this way, we can empirically evaluate if the ( β, δ ) model predicts the characterisation of people according to their time preferences from the present bias and the discount factor, in the way argued by O’Donaghue and Rabin (1999, 2008). 3. Methodology and database 3.1. Procedure Our study is founded on a database of degree students that we elaborated ourselves and which was obtained via surveys. Its design, described below, analyses their behaviour in planning, preparing and developing of their academic activities, as well as their results. This behaviour differs among them in the planning of the preparation of the subject and in the degree to which they fulfil this plan. Furthermore, within the group of those who do not 103 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation fulfil their plan, people can be distinguished by their degree of awareness of their future failures. The students likewise differ in their intertemporal preferences, characterised mainly by each one’s discount factor and present bias. A time discount task was implemented to infer the discount rates and the possible present biases. On the other hand, in order to classify the people we had to obtain information on the extent to which they fulfil their plan and if they are aware, a priori, of what they are going to do. The questionnaires that include the experimental discount task have been designed combining the methodologies developed by Burks et al. (2012) and by Wong (2008). The hypothesis which we verify is that the likelihood of the students fulfilling their plans depends on the type of intertemporal preferences that they have, controlling for their different personal and socio-economic characteristics. Moreover, we examine if the present biases lead to self-control problems and, where appropriate, to supporting the costs of well-being which they cause, as O’Donaghue and Rabin (1999, 2008) predict. In this way we test the empirical basis of their model. The sample includes students of the compulsory subject of Macroeconomics, corresponding to the second course-year of the Degree of Finance and Accounting of the University of Seville, Spain. The subject is taught in 8 groups of a similar size: half in the morning and the other half in the afternoon. The centre determines the assignation of the students to the groups by extra-academic criteria, which establishes a similar profile in all of them. The content of the course is the same and includes an identical exam for all the students. 3.2. Questionnaires and classification of individuals The data was obtained through two questionnaires. The first questionnaire is done midterm and contributes most of the information, including the experiment and the controls. At this point in the course, the students had information on the content of the subject and its difficulty and could carry out a precise estimation of the requirements of the work needed to prepare it. The survey was carried out during the same week with all the groups in the second hour of a two-hour session. Each student signed his/her authorisation to participate in the experiment and read the paper’s instructions. The instructions clarified the voluntariness of the activity and that it would not affect the mark of the subject in any way. Likewise, they indicated precisely how to do the time discount task. The process by which the data would be anonymised was also explained, placing special emphasis on the need to give sincere answers. In this way, the students had incentives to respond to the discount task. This made it possible to also sincerely answer the rest of the questions. The researchers did not teach most of the groups and significant differences were not found in the answers given by their students and the rest. In any case, in spite of the insistence on the sincerity of the answers and that there was an anonymous handling of the questionnaires from the moment in which they were handed out, there exists the possibility of an experienced demand effect, as authors such as Zizzo (2010) and De Quidt et al. (2017) indicate. Nevertheless, these same authors 110 david patiño and francisco gómez-garcía include all the controls. The result shows that the likelihood of being consistent decreases the greater the present bias is. This agrees with our initial hypothesis, though the relation is relatively weak. The complete model (4) also identifies other characteristics of people with consistent time preferences. Specifically, they tend to do a greater total number of study hours. In contrast, students with lower levels of monthly allowances have a lower tendency to have time preferences of this type. This result suggests that students with lower levels of income could be adopting worse decisions, which coincides with recent findings in this vein15. Table 3 RELATIONSHIP BETWEEN CONSISTENT PEOPLE AND THEIR TIME PREFERENCES Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Delta -7.107 -6.601 -8.901 -8.846 (7.250) (7.302) (7.058) (7.085) Beta 0.238 0.232 0.273* 0.278* (0.170) (0.172) (0.163) (0.163) Monthly allowance under €200 -0.127* -0.115* (0.067) (0.067) Monthly allowance between €200 and €300 -0.135* -0.129* (0.073) (0.075) Total study time 0.002*** 0.002*** (0.000) (0.000) Constant 7.002 6.920 8.957 8.600 (7.122) (7.179) (6.902) (6.939) Observations 214 214 209 209 R-squared 0.009 0.019 0.235 0.250 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The analysis has been limited to subgroups in order to better know the characteristics of people with these preferences. Specifically, the relation between the β factor and the condition of consistent is only found when the sample is limited to women, to students of the afternoon classes and to non-repeaters. In the case of the first two groups, the significance rises to 1%. A relation between the present biases and consistent preferences is not noted in the rest of the groups. This indicates that there may be different patterns which explain people’s type of time preferences. The robustness of the results has been tested estimating logit and probit models. The estimations are similar to those obtained with the LPM, though the t statistic of the β factor coefficient only attains the value of 1.32 and cannot be considered statistically significant. 111 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation Table 4 shows the value of the coefficients and the standard deviation of the δ discount rate and of the β factor of the estimations of the groups considered and of the discreet choice models for the model which includes all the controls. Table 4 RELATION BETWEEN CONSISTENT PEOPLE AND THEIR TIME PREFERENCES. ANALYSIS OF THE ROBUSTNESS OF THE RESULTS Variables (1) Probit (2) Logit (3) Men (4) Women (5) Repeaters (6) Non-repeaters (7) Morning classes (8) Afternoon classes Delta -23.256 -39.977 -2.029 -12.542 -15.586 -9.509 -1.837 -11.850 (52.058) (95.744) (11.465) (9.352) (18.671) (8.247) (11.622) (10.031) Beta 1.416 2.752 0.093 0.559*** 0.213 0.339* -0.235 0.770*** (1.133) (2.088) (0.290) (0.205) (0.331) (0.199) (0.242) (0.239) Constant 14.931 20.211 2.990 11.463 15.294 9.258 2.628 10.962 (453.614) (899.217) (11.199) (9.215) (18.687) (8.066) (11.414) (9.741) Observations 209 209 89 120 44 165 104 105 R-squared 0.303 0.422 0.436 0.281 0.280 0.384 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The rest of the time preference typologies do not have a statistical relation with the present biases. The specific case of the naïve is especially outstanding as the theory considers that the disproportionate preference for immediateness is its main determinant. Tables 5 and 6 show the inexistence of a statistical relation between discount rates and present biases and the consideration of people as naïve or sophisticated. Table 5 RELATIONSHIP BETWEEN NAIVE PEOPLE AND THEIR TIME PREFERENCES Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Delta -0.385 -0.643 0.145 -0.232 (8.519) (8.568) (9.059) (9.151) Beta -0.008 -0.010 0.010 0.004 (0.200) (0.201) (0.209) (0.210) Age-squared -0.001 -0.001* -0.001 (0.000) (0.001) (0.001) Constant 0.567 0.140 -0.621 -0.073 (8.369) (8.424) (8.858) (8.962) Observations 214 214 209 209 R-squared 0.000 0.013 0.058 0.065 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 112 david patiño and francisco gómez-garcía Table 6 RELATIONSHIP BETWEEN SOPHISTICATED PEOPLE AND THEIR TIME PREFERENCES Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Delta -3.962 -4.195 -10.191 -11.262 (11.099) (10.997) (11.291) (11.400) Beta -0.076 -0.051 0.028 0.039 (0.261) (0.258) (0.261) (0.262) Age -0.088** -0.084** -0.081* (0.040) (0.041) (0.042) Age-squared 0.001* 0.001* 0.001* (0.001) (0.001) (0.001) Gender -0.107 -0.152** -0.167** (0.069) (0.072) (0.074) Living with their parents -0.195*** -0.196*** (0.072) (0.072) College-educated father 0.239** 0.260** (0.101) (0.103) College-educated mother -0.225** -0.225** (0.106) (0.107) Constant 4.464 6.078 12.078 13.023 (10.903) (10.812) (11.041) (11.166) Observations 214 214 209 209 R-squared 0.003 0.045 0.144 0.151 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. As is noted in Table 6, the students of the sample who are women, those who do not live with their parents and the youngest have a greater tendency to behave as sophisticated. No characteristic is found which favours the consideration of people as naïve. The robustness checks carried out confirm these results with the unique exception of men with naïve preferences. This group has a tendency to present lower discount rates. That is to say, the most impatient men are more likely to have naïve preferences but, contrary to what the theory suggests, this probability decreases with the present bias (it grows with the β factor) in a marginally significant manner. Table 7 shows the estimated values for naïve students in different groups of the sample. As we indicate, the tendency to procrastinate is also measured through a subjective index of postponement of academic tasks. Table 8 estimates the relation between this and time prefer- 113 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation ences through the LPM. As we see, the degree to which people procrastinate is not explained by the parameters which determine their time preferences. The total number of hours of study for the exam and the students’ satisfaction with the career studied are the only determinants of their propensity to procrastinate. That is to say, the more motivated and more studious students are the ones who tend to postpone their tasks less. In spite of the relations not being statistically significant, the important quantitative effects that the discount rates and the present biases have also stand out, considering the small variations which are produced in these variables. Table 7 RELATION BETWEEN NAÏVE PEOPLE AND THEIR TIME PREFERENCES. ANALYSIS OF THE ROBUSTNESS OF THE RESULTS Variables (1) Probit (2) Logit (3) Men (4) Women (5) Repeaters (6) Non-repeaters (7) Morning classes (8) Afternoon classes Delta 0.210 -10.522 -30.654** 17.081 -0.209 -8.907 -18.235 5.429 (36.970) (64.152) (14.299) (12.981) (38.391) (9.547) (14.387) (14.058) Beta 0.042 0.273 0.678* -0.301 -0.303 0.295 0.464 -0.328 (0.834) (1.480) (0.362) (0.284) (0.680) (0.231) (0.300) (0.335) Constant -8.409 -4.889 29.145** -16.063 7.031 7.801 16.807 -4.791 (35.700) (61.869) (13.967) (12.791) (38.423) (9.338) (14.130) (13.652) Observations 196 196 89 120 44 165 104 105 R-squared 0.240 0.122 0.411 0.125 0.156 0.115 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The analysis of the robustness of this variable does not include the estimation of discrete choice models given that we have treated it as continuous. Table 9 shows the rest of the groups considered before. As we can see, a statistically significant relation between the discount rates and the procrastination index has been found in the case of the men, the non-repeaters and the students in the morning groups. In these groups, the people with greater discount rates tend to postpone their academic activities to a greater extent than the rest. This reveals a relation between impatience and procrastination. The explanation of procrastination through the quasi-hyperbolic discount also predicts that people with high present biases tend to have self-control problems as they postpone the tasks that they plan to carry out. In our case this can lead to the students not fulfilling the plan of preparation for the exam and finishing up with bad results. Therefore, there should exist a relation between the academic results and the present biases. To check this, we have estimated a model similar to the previous ones, but which controls by the group in which the student attends class in order to group together the differences in the teacher, the companions and the class hour. Additionally, variables related to the activity and the educational training of the parents have been introduced. The third group of variables has been maintained as in the rest of the estimations. 114 david patiño and francisco gómez-garcía Table 8 RELATIONSHIP BETWEEN THE SUBJECTIVE INDEX OF POSTPONEMENT AND TIME PREFERENCES Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Delta 52.510 67.667 66.617 54.050 (70.122) (69.194) (71.373) (70.154) Beta 0.047 -0.364 -0.690 -0.545 (1.642) (1.621) (1.642) (1.610) Age 0.396 0.511** 0.335 (0.248) (0.257) (0.257) Age-squared -0.008** -0.009** -0.006 (0.004) (0.004) (0.004) Total study time -0.009*** -0.009*** (0.003) (0.003) Satisfaction with their studies -0.453*** (0.156) Constant -44.760 -64.660 -61.378 -43.958 (68.916) (68.012) (69.806) (68.743) Observations 211 211 206 206 R-squared 0.005 0.053 0.134 0.184 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Table 9 RELATIONSHIP BETWEEN THE SUBJECTIVE INDEX OF POSTPONEMENT AND TIME PREFERENCES. ANALYSIS OF THE ROBUSTNESS OF THE RESULTS Variables (1) Men (2) Women (3) Repeaters (4) Nonrepeaters (5) Morning classes (6) Afternoon classes Delta 213.095* 14.602 -86.065 136.019* 189.116* 78.862 (119.312) (91.503) (209.734) (75.653) (102.543) (106.190) Beta -5.760* 0.403 -0.637 -1.972 -0.479 -2.013 (2.956) (2.030) (3.714) (1.832) (2.111) (2.529) Constant -191.504 -11.644 66.884 -120.056 -180.452* -55.253 (116.759) (90.124) (209.740) (74.029) (100.721) (103.124) Observations 87 119 43 163 101 105 R-squared 0.379 0.177 0.537 0.285 0.295 0.304 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 115 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation As can be seen in Table 10, there is a positive relation between the β factor, the inverse of the present bias and the marks. Additionally, the negative sign of the discount factor indicates that as it grows (the lower the discount rate), the marks decrease or, in other words, people with a lower preference for immediateness or with a greater tendency to wait obtain better scores. Both results are coherent with the intuitive idea and verify the hypothesis predicted by the theory. However, no statistical relation has been found between the different types of students and their marks. Table 10 RELATIONSHIP BETWEEN MARKS AND TIME PREFERENCES Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Delta 10.216 -26.666 -93.501** -91.110** (40.998) (38.622) (40.161) (40.140) Beta 0.123 0.945 1.956** 1.967** (0.965) (0.916) (0.950) (0.943) Gender 0.353 0.542** 0.444* (0.243) (0.260) (0.268) Group 2 1.995*** 1.385** 1.245* (0.623) (0.667) (0.679) Group 3 2.667*** 2.063*** 1.886*** (0.597) (0.654) (0.655) Group 4 1.579** 0.892 0.923 (0.612) (0.660) (0.658) Group 5 2.738*** 1.735** 1.679** (0.624) (0.680) (0.683) Group 6 1.416** 0.539 0.394 (0.623) (0.681) (0.682) Group 7 0.456 -0.338 -0.376 (0.624) (0.681) (0.678) Group 8 1.732*** 0.652 0.502 (0.665) (0.724) (0.723) Monthly allowance between €200 and €300 -0.706 -0.725* (0.430) (0.430) Living with their parents -0.610** -0.578** (0.259) (0.257) Degree of risk 1 2.360** 1.967* (1.117) (1.123) Degree of risk 2 2.826*** 2.417** (1.056) (1.065) 116 david patiño and francisco gómez-garcía (Continued) Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Degree of risk 3 2.433** 1.903* (1.024) (1.045) Father entrepreneur +10 employees 0.756 0.876* (0.520) (0.523) Mother pensioner -1.405** -1.507** (0.630) (0.633) Mother without education -0.878* -0.760* (0.451) (0.452) Mother self-employed worker -1.793*** -1.677*** (0.478) (0.477) Satisfaction with their studies 0.214** (0.095) Constant -3.558 30.611 95.599** 91.168** (40.277) (37.920) (39.208) (39.292) Observations 206 206 187 187 R-squared 0.001 0.227 0.395 0.416 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The controls indicate that the men obtain better marks. There exists a group effect –that is to say, either the teacher, the companions or the class time– which generates consequences in the marks. The students who live with their parents have worse results. It may be that their productivity has less stimuli as they have less direct costs. The motivation which studying a satisfactory career causes is also transformed into better results. Lastly, the children of pensioner mothers, those of self-employed workers or, those without studies, get worse marks. The last implication for the model that we are going to test is the prediction that selfcontrol problems generate well-being costs in the people who suffer from them. According to this, not fulfilling the objectives planned causes dissatisfaction with one’s own actions. Therefore, we measure the degree to which utility and decision distance themselves from each other and if the divergence is brought about by time preferences. To test this implication, the relation between the factors that determine β and δ is estimated and the level of subjective well-being is measured through satisfaction with life in general. Table 11 shows the results. As can be seen, the discounting factor has a positive effect upon the degree of satisfaction with life, having a statistical significance at the 5% level. According to this, more impatient people would be, removing the rest of the elements considered, those who report 117 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation greater levels of satisfaction. The present biases do not have a statistically significant relation with the rates of happiness16. In this sense, and considering that this factor is the main determinant of the self-control problems, the evidence found in the sample of students analysed does not corroborate this implication of the theory. The other factors which favour satisfaction with life are self-confidence, satisfaction with studies and, trust in others. Having a rural origin and risk aversion also have a positive and significant effect. Table 11 RELATION BETWEEN THE LEVEL OF SUBJECTIVE WELL-BEING AND TIME PREFERENCES Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Delta 52.267 59.246* 53.580 53.074** (35.979) (35.215) (34.540) (26.809) Beta -0.499 -0.574 -0.512 -0.336 (0.835) (0.817) (0.798) (0.617) Age -0.236* -0.195 -0.021 (0.125) (0.126) (0.099) Monthly allowance under €200 -0.886*** -0.622** (0.326) (0.255) Monthly allowance between €200 and €300 -1.034*** -0.862*** (0.359) (0.283) Monthly allowance between €300 and €500 -1.014** -0.900*** (0.443) (0.342) Degree of risk 1 1.962** 0.943 (0.973) (0.764) Degree of risk 2 2.520*** 1.726** (0.911) (0.717) Degree of risk 2 2.687*** 1.612** (0.896) (0.714) Total study time 0.002* 0.002 (0.001) (0.001) Trust in others 0.111** (0.053) Satisfaction with their studies 0.224*** (0.060) Self-confidence 0.405*** (0.047) Constant -44.104 -46.909 -43.634 -49.765* (35.349) (34.605) (33.776) (26.258) 118 david patiño and francisco gómez-garcía (Continued) Variables (1) model 1 (2) model 2 (3) model 3 (4) model 4 Observations 213 213 209 209 R-squared 0.011 0.078 0.213 0.539 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 5. Conclusions The consequences of biases in behaviour and problems of self-control have become important in economic analysis and are one of the bases of behavioural economics. Among the most accepted explanations of self-control problems it is noted that people’s time preferences follow a quasi-hyperbolic form instead of the traditional discount. According to this, the absence of self-control is the consequence of the effect that present biases or a tendency to shortsightedly value the events which take place in the present have, overestimating them compared to those that will happen in the future. The basic interpretation of this proposal means that the people who have a disproportionate preference towards the present adopt different decisions when the moment of making a postponed decision approaches – and that this signifies assuming costs – from those that they would have made if they had decided in advance. The bias leads them to modify their cost-benefit analysis and to adopt decisions far removed from their long-term preferences. Under these premises, O’Donaghue and Rabin defined in various articles a classification of people which makes them more or less inclined to suffer self-control problems. These categories are mainly explained by the existence of present biases. Some of these types of people, specifically the naïve, do not manage to plan their behaviour or, seen from another perspective, repent, a posteriori, about decisions adopted in the past. The proposition is attractive and articulates a logical explanation of this phenomenon but opens the door to new conceptual difficulties. For example, the economic analysis is based on people with stable preferences, at least during the time that the analysis lasts. This stability is the logical consequence of supposing rational people who do not randomly change their behaviour and who can therefore be the object of prediction. Admitting the possibility of present bias implies recognising that individuals can change their preferences or question the rationality of their behaviour. This has far-reaching implications for the ways of reasoning of economists. The quasi-hyperbolic formulation has been considered in the literature of behavioural economics, which uses it extensively to model time preferences that underestimate the future (or potential) utility in relation to the current utility due to present biases. In brief, it constitutes a form of measuring the degree to which the decisions of the agents distance themselves from ideals and, therefore, opens the way to measuring the need of a public action. In general, to calculate the degree to which the real behaviours of people are removed from their 119 Do Quasi-Hyperbolic Preferences Explain Academic Procrastination? An Empirical Evaluation predictions through conventional models is a way of determining if policymakers should influence their actions. Nevertheless, the main shortcoming of the approach is that the empirical verification is limited and, in any event, a good part of its logical developments is partial. The present article presents an unprecedented empirical verification of different aspects of the explanation and of the relationship between discounting rates, present-bias and self-control problems, carried out with a sample of university students during their activity. Specifically, it empirically checks the determinants of the different types of temporal preferences as well as their consequences. The preparation for the final exam of a subject is an example of a costly activity that is carried out in the future, susceptible of being planned for and therefore a candidate is likely to be affected by self-control problems. And though we have centred on the effects of present bias on the actions of university students, the approach can be extended to other topics. For example, Paserman (2008) estimates an employment search model with similar premises and uses it to predict the effect of the policies of unemployment benefits. Our results show relatively weak evidence of the approach. Specifically, we have found empirical evidence that the lack of present biases leads to consistent behaviours, or, in another words, people who tend to fulfil the time plan that they had imposed on themselves. Furthermore, the relation only occurs for some groups of the sample, not for all, although in these groups it is sufficiently strong to be applied to all the sample. However, we have not found a clear empirical guarantee to the rest of the explanation’s implications. We have especially not encountered a relation between the present biases and the time preferences of the people who behave as naïve or sophisticated. We have only identified a relation between discount rates and naïve preferences in men. Nor do our calculations enable corroboration of the majority of the consequences that the ( β , δ ) model predicts of self-control problems. Though we have noted that the absence of present biases improves academic results, a statistical relation has not been established with the levels of satisfaction with life that we have used as a measure of the self-control problems’ costs of wellbeing. Our results must be evaluated very cautiously. To take due account of them, it is necessary to recognise that the mechanism that underlies financial behaviour is not, necessarily, the same as that of another type of decision and to be aware that the behaviour in preparing an exam has been inferred through a hypothetical questionnaire. All this implies the need for additional research to be done to explore if the predictions of the model fail to be confirmed because of a failure of the theoretical model itself or for alternative reasons. Nevertheless, these limitations do not invalidate the main conclusion of our work, which shows that even recognising that the ( β , δ ) model represents a logical solution to explain self-control problems, it is necessary to analyse its empirical basis to be sure that it is an appropriate explanation of the behaviour of people, which may serve as a guide for possible proposals of public action.