Migration has been a major source of human survival, adaptation, and economic change over centuries across regions. Today, more than ever, migration forms a central part of the global flow of persons, goods, practices, and ideas. Though intellectual discussions on human mobility is bit leaning towards mobility across nations, especially from developing to developed countries, but interestingly most movement in the world does not take place between countries. The overwhelming majority of people who move do so inside their own country. Data reveals that internal migration is four times higher than international migration  . And when it comes to internal migration, case of Indian economy is of particular interest because of its heterogeneity in terms of socio-economic and demographic characteristics. Though historically migration in India has not been very prominent due to various socio-economic and political constraints (Davis 1951), but the picture has changed in recent times. Due to rapid transformation of Indian economy along with other factors, internal migration has got a rapid motion. According to Census of India report, inter-state migration based on the place of last residence criterion has increased from 159.6 million in 1971 to 201.6 million in 1981 and then to 225.9 million in 1991 thereby registering a decadal growth of around 19 percent. But the growth has been most prominent in the post 90s. In 2001, figure of internal migration has reaches 309.3 million marking more than 37 percent growth in the decade  . Recently, this increasing trend of migration has raised some serious concerns in regard to its effect on social security and crime  . In this paper, we made an attempt to examine the linkage between interstate migration and crime in India.
Inter-State Variation of Crime in India
Since crime has many faces so it becomes very difficult to analyse the relative position of a region in terms of crime in general. In case of India we have more than half a dozen of violent crime and incidentally these crimes do not move in the same direction across regions. For instance, Karnataka ranks top in terms of murder rate while Gujarat stays at the bottom. Similarly, rate of rape has been found to be maximum in Mizoram and minimum in Pondicherry. Rate of kidnapping & abduction has been observed highest in Delhi (Crime in India, 2009)  . Thus, in order to analyse regional variation of crime, we have calculated a composite crime index for each of the major fifteen states of India for three time points: 1991, 2001 and 2009. This Index has five dimensions: (a) murder and attempt to commit murder, (b) rape, (c) kidnapping and abduction, (d) dacoity and robbery, and (e) riots  . Index for each of these dimensions is first calculated using the following formula:
Where CIij stands for crime index of ith dimension for jth state, Xij for actual value of ith crime rate in jth state, Min(Xi) and Max(Xi) stand for maximum and minimum values of ith crime rate during the period from 1991 to 2009. Each of these dimension indices has a numerical value between 0 and 1. The scores for these five dimensions are then averaged in order to get the composite index for crime:
Table1 summarises this composite crime index for the fifteen major states.
Table 1 about here
In 1991, CI value varied from a lowest value of 0.150 in Kerala to a highest value of 0.571 in Bihar. Similarly in 2001 the lowest and highest figures were 0.068 (West Bengal) and 0.448 (Assam) respectively. And in 2009, Gujarat witnessed a lowest CI value of 0.130 and Assam retained its position at the top securing 0.520 for its CI. In the early 90s, Bihar, Madhya Pradesh, Assam, Rajasthan and Uttar Pradesh were among the states with higher CI. States like Kerala, West Bengal, Haryana, and Tamil Nadu were at the other end. However, the picture has changed over the time. Bihar which was at the top in 1991 comes down to 5th rank in 2009. Similarly, Uttar Pradesh which ranked 4th position comes down to 14th position. On the other hand, states like Orissa and Haryana who were relatively in the bottom position climbed up in the list. On the whole, Assam, Madhya Pradesh, Bihar and Rajasthan have been catagorised as the most crime-prone states in the nation; while West Bengal, Tamil Nadu, and Andhra Pradesh have been grouped in the list of less crime prone. Moreover, the gap in CI index between top and bottom state which was 0.421 in 1991 came down to 0.380 in 2001 and marginally increased to 0.390 in 2009.
Human Mobility and Crime: Theoretical Framework
Literature on crime-migration nexus is largely ambiguous. While some scholars consider migration as a potential source of crime, others believe that migration is a potential source of development and does not necessarily fuels criminal activity. It has been argued by former group of scholars that migration leads to demographic transitions which acts as stimulator of criminal activity in a society. There are two variants of the demographic transition framework. The first argument is compositional in nature. According to this argument, crime follows a distinctive age pattern with criminal activities being highest among teens and young adults (Hirschi and Gottfredson 1983). And also involvement of males in crime is relatively higher than that of females. Thus, to the extent that migration increases the percentage of young and male population in the destination society, crime rates will increase. The second demographic transition argument is contextual in nature which says that crime rates will rise when rapid social change breaks down the social networks and institutions necessary for effective socialization and behavioural regulation. One change believed to contribute to social disorganization is population instability. As a major driver of population change and residential instability, in-migration may thus be regarded as a critical factor behind the breakdown of informal social control and concomitant increase in crime rates (Bankston 1998; Lee, et al. 2001; Lee and Martinez 2002; Mears 2001; Reid et al. 2005). Another argument against migration steams from realisation of economic deprivation. If migrants are mostly unskilled and lack modern way of working then it will reduce their job prospects. As a result, they will expose to unemployment, poverty, and other social ills. Migrants may thus feel that opportunities for attaining economic success through legal paths are bleak. According to opportunity structure theory, that realisation can lead to strain and frustration, which will increase the probability of adaptive responses that involve alternative economic pursuits, such as crime (Lee et al. 2001; Mears 2002; Reid et al. 2005). But the argument that migrants lack human capital and hence bear the risk of being involved in criminal activity has not been accepted by others. There are scholars who argued that migrants are not necessarily a random cross-section of the source society but are a self-selected group with relatively higher level of potentiality, aspiration and low criminal inclination (Butcher and Piehl 2005). Similarly, Ousey and Kubrin (2009) extended an interesting hypothesis on negative association between migration and crime. They argued that sometimes concern among the general public regarding unpleasant impact of migration might put pressure on governments to strengthen the formal social control machinery, which in turn contributes to decrease crime and violence in the society. There is still another school of thought who believes that it is crime that actually influences people to migrate. Thus, there is no reason to think that migrants are necessarily a sample drawn from criminals. Using public opinion surveys of approximately 49,000 respondents, Wood et al. (2010) found that the probability of seriously considering family migration to the United States was around 30 percent higher among respondents who reported that they or a member of their family was a victim of a crime sometime during the year prior to the survey. Therefore, crime-migration linkage is very complicated and multi-directional. The nature of linkage between crime and migration should not be pre-assumed rather it should be examined.
Model Specification and Data Source
Given the theoretical background in last section, here an attempt has been made to specify the model. Since our dependent variable is an index whose value ranges from 0 to 1, therefore we have specified our model in the following non-linear form to avoid the unboundedness problem  :
Here, CIit stands for crime index in ith state in year t, NMRit-1 for migration rate in ith state in year t-l , Âµt for random error term in year t, and Î±s are coefficients of the model. We have used lags of the explanatory variables since literature suggests that today's crime is caused by yesterdays push and pull factors  . If coefficient of NMR (ß1) found to be statistically significant then it would mean that migration has influence on crime. The nature of this influence, however, depends on the sign of ß1. Note that this specification also contains a vector X/ (in lag form) of control variables which are supposed to influence crime and the respective vector of associated coefficients ß/.
The literature is replete with colorful terms to hypothesize variables other than migration that might affect crime. For example, Ehrlich (1973) argued that the probability of a person being involved in criminal activity depends on economic risk of legal activity. If risk of legal activity is high, as reflected by high poverty rate, then there is greater possibility of criminal activity. Therefore, one should expect positive association between poverty and crime. One can also theorise the relationship between poverty and crime in somewhat different way. Since marginal utility for income is relatively higher for poor people therefore it is relatively easier to put them into an illegal activity or make them economically forced to commit a crime. Besides, poor people have relatively less information regarding the consequences of crime because of poor education and less access to media, which make them yet more vulnerable (Kelly 2000). Urbanization is often regarded as an important driver of crimes. People living in rural communities are more likely to know one another socially than people living in large cities. The degree to which members of the community know each other is termed 'density of acquaintanceship (Weisheit and Donnermeyer, 2000). It has been suggested that a high density of acquaintanceship leads to increased watchfulness of citizens (informal surveillance), which in turn leads to lower rates of crime (Freudenburg, 1986). The reverse is true for urban areas. Because of high density and crowding, criminals in urban areas have better chance of hiding themselves hence fewer odds of being arrested and recognised. Therefore, it is argued that as urbanisation increases incidence of crime also increases(Glaeser et al 1996; Galvin 2002; Gaviria and Pagés 2002). But this inverse relation between urbanisation and crime is far from accepted by all scholars. Barclay et al (2004) suggest that high cohesiveness within communities, as happens in rural areas, may actually enable the commission of some crimes due to increased tolerance and 'techniques of neutralisation' whereby people develop a set of justifications that allow them to excuse their or their neighbour's deviant behaviour. Moreover, rural areas are basically poor in terms of anti-crime institutions and infrastructures and so, one can expect relatively higher incidence of crime in rural areas. Therefore, the urbanisation-crime nexus is not very clear and conclusive; this is rather an empirical issue. Yet another factor that might have an impact on crime is education. There are a number of reasons to believe that education can reduce criminal activity. First, education increases the returns to legal work, raising the opportunity costs of illegal behaviour. Second, education directly affects the financial or psychic rewards from crime itself. Finally, education may alter preferences in indirect ways, which may affect decisions to engage in crime (Lochner and Moretti, 2001). Since crime is supposed to be prevented by formal social control machinery, therefore a society with higher strength of police force is expected to have lower occurrence of criminal activities.
Specifying the control variables, equation (1) can be re-written as:
where AL, PCP, Urb and P stands for adult literacy rate, proportion of civil police, urbanisation, and poverty ratio. These are all control variables and supposed to affect crime some way or other  .
Equation (2) represents what is known as the (cumulative) logistic distribution function.
If Cit is given by equation (2), then (1-Cit ) is
Dividing (2) by (3), we write
Taking natural log of (4), we obtain
where Zit stands for log of odds ratio of crime index. It is interesting to note that although we start with a non-linear model, our final model (6) becomes not only linear in variables but also in parameter. However, measure the change in Z for a unit change in independent variable, that is , it tells how the log-odds of CI (rather than absolute CI) change as independent variable changes by one unit. The intercept is the value of the log-odds when value all explanatory variables attain zero. Like most interpretations of intercepts, this interpretation may not have any physical meaning.
We have considered 15 major states which possessed all the relevant data for analysis  . These states are in the North, South, Central, East, West and North-Eastern zones of India. Data have been taken from the public domains of governmental and related organizations. Details of these are given in Appendix A.
Results and Analysis
We have estimated three different equations by adding or subtracting one or more of the explanatory variables while keeping the dependent variable (log of odd ratio of crime index) unchanged. Estimation method of each equation has been decided through formal test. Table 2 summarises the estimated results.
Table 2 about here
Before we analyse the results, it is important to note that the null hypothesis of homogeneous intercept across the state economies is rejected in the first equation at the conventional level of significance  . This justifies use of fixed effect panel model for this equation which allows for state heterogeneity. Also note that the value of Adj. R-square which ranges from the lowest of 0.13 to the highest of 0.26 is quite satisfactory given the panel nature of regression. The F-statistic in all the equations is significant at beyond the one percent level, attesting to the overall strength of the model. Turning to analysis, column (1) presents the base specification, in which we have net migration rate as the sole regressor. Columns (2) and (3) present alternative specifications that examine effect of migration on crime while controlling other factors supposed to influence crime.
We can see that net migration rate proved insignificant in all equations, indicating that there is no significant association between migration and crime. This result immediately nullifies the popular impression in the nation that internal migration is responsible for increasing crime in the nation. Proportion of civil police force, as expected, has a significant negative effect on crime in all equations. Similarly, urbanisation and adult literacy rate also proved significant with negative sign for their coefficient in the second equation. This implies that both urbanisation and adult literacy rate have negative effect on crime. We attribute this to the impact of better anti-crime institutions and higher probability of criminals being detected in urban areas, and secondly, to the positive impact of education on legal work, raising the opportunity costs of illegal behaviour. Finally, in column (3), we see that coefficient of poverty ratio is positive and highly significant. This result, therefore, justifies the hypothesis that there is a positive linkage between poverty and criminal activity. However, it is interesting to note that when poverty ratio is added in the equation, both adult literacy rate and urbanisation become insignificant, though sign of their coefficient remains meaningful. This implies we cannot trace out the independent impact of literacy and urbanisation on crime keeping poverty ratio constant. This suggests that much of the impact of adult literacy rate and urbanisation on crime appears to occur through the linkage between poverty, education and urbanisation.
In this paper, we have examined the linkage between interstate migration and crime in India. Our analysis shows that the association between these variables is not significant. Therefore, the study discards the controversial thought that migration is responsible for increasing crime in the nation. Government must understand this truth and should think for some constructive ways to control crime, rather than staring at migration. We have, however, identified some factors which were found to have significant association with crime. Economic deprivation as captured by poverty ratio has a significant positive impact on crime, while factors like proportion of civil police, adult literacy rate, and urbanisation all have negative impact on crime. Thus, the obvious policy suggestion that appears is that government should concentrate on removal of economic deprivation by opening up more employment ventures to weaken the pull factor behind crime in the nation. To strengthen the push factor, police system can play a great role, especially the civil police. Ongoing process of raising the level of education in the nation should be continued and made more value inferring. We strongly believe that education can go a long way in preventing criminal activity by increasing the returns to legal work and by raising the opportunity costs of illegal behaviour.