Visualizing timescape issues in a comparative study of the American states

Introduction

The American states are sometimes called laboratories of democracy. This characterization is more normative than empirical, but it does reflect an important ongoing and underappreciated element of American politics: given the opportunity to be different, the states are different. Using the laboratory analogy, states attempt to deal with common social problems by pursuing different policies. The ultimate goal is to identify empirical relationships between policies and problems that are causal and use this intelligence to identify and pursue more effective policies. We do not have the experimental control over treatment and control groups we might prefer, but we do have the opportunity to study governments using the comparative method.

The states are excellent candidates for comparative analysis over time using comprehensive data. They are key units of government in the American federal system – they created the federal government and they create local governments. They share a federal constitution and considerable history. They all seek to improve the lives of their residents through effective government policies. They pursue quite different policies to deal with the challenges they have in common.

The American states have been the subject of quantitative empirical analysis for more than six decades. Thousands of social scientists and their students have studied the states using data from multiple years. Their publications have dealt with time issues in almost every way imaginable, including ignoring them. ¹ Study of the American states could benefit from insights in the timescape literature that views time as an important resource and analyzes the roles time plays in the political process. ²

Recently, the most common research design has combined state data for multiple years and used sophisticated constant-coefficients cross-sectional time-series statistical models to parse out relationships of interest. ³ However, published articles discussing methodological issues seem to equal or outnumber more substantive studies. The methodologists seem to disagree more than they agree with each other. ⁴

At the same time, some have pursued simpler analyses (Luttbeg, 1992, 1999, 2005). As a result, we know quite a bit about empirical relationships across the states and over time. We know that some are consistently strong year after year and others only exist at some points in time. Inconsistent empirical relationships are sufficient grounds for rejecting models that posit consistent relationships. Stable measures of association are necessary but not sufficient to document that models stand the test of time.

All who study quantitative empirical data have something in common. At least once in our careers we were warned not to rely solely on summary statistics such as measures of association and model parameter estimates. We were encouraged to review graphics such as scatterplots to check whether important assumptions are met, to identify outliers and to recognize patterns not apparent from statistics. In contemporary terminology, this advice is: visualize data as part of the model evaluation process (Tufte, 1983). Another thing we have in common is that we almost always ignore this advice.

Anscombe illustrated the importance of visualizing data most vividly (Anscombe, 1973). He presented four data sets of two variables with identical numbers of observations, mean of the x values, mean of the y values, sums of squared errors (about the mean), regression sums of squares errors (variance accounted for by x), residual sums of squared errors (about the regression line, correlation coefficient and coefficient of determination. Moreover, each data set had an identical equation of the least-squared regression line. Yet, as the scatterplot visualizations in Figure 1 demonstrate, the four data sets are strikingly different. OPEN IN VIEWER

According to Anscombe (1973), his purpose was merely to suggest that graphic procedures are useful. One of the purposes of this article is join others who have suggested that graphic procedures using animation to show change over time are useful (Battista and Cheng, 2011). Animations can help us to recognize and understand better temporal analysis phenomena such as temporality, timing, time frames and periodization, patterns, tempo, duration, tempo, duration, sequence and critical junctures (Adam, 2004; Howlett, 2009; Meyer-Sahling and Goetz, 2009; Grzymala-Busse, 2011).

The animated data visualizations presented here were created using the Motion Chart Gadget available through Google Documents. Battista and Cheng (2011) identify a number of other options to create similar visualizations, including Gapminder World, Google API, Tableau Public Web, JMP by the SAS Institute and Trend Compass. The Statistics Online Computational Resource at the University of California Los Angeles has an online motion chart builder at socr.ucla.edu, and Microsoft Excel 2013 also has the capacity to animate scatterplots to show change over time.

Issue One: Proactive and reactive policy

Ideally, the goal of government action is to reduce problems and enhance the lives of citizens. That leads us to assume and expect the following causal direction: effective government actions cause better social conditions. When social condition is measured in terms of problem size, there should be a negative correlation with government action. When social condition is measured in terms of desired outcomes – successes and benefits – there should be a positive correlation with government action. Government is assumed to be proactive in dealing with problems and working to achieve goals and viewing government as the actor is appealing. On reflection, the opposite view is equally plausible: government is reactive, not proactive.

If we always view state governments as powerful, effective, proactive agents, we always expect that state program efforts will be positively correlated with good results. If we always view state governments as reactive agents, we always expect that state efforts will be negatively correlated with good results. A more sophisticated approach would be to recognize occasionally state governments are powerful actors and occasionally they are reactors whose ability to improve bad situations is limited.

Although he did not distinguish proactive from reactive government efforts explicitly, we can see clear evidence in Ira Sharkansky’s study more than 35 years ago (Sharkansky, 1967). He tested the idea that state and local governments that spend more have better outcomes than state and local governments that spend less, by calculating more than 200 correlation coefficients that interrelated spending and outcome measures. Only 38% met his threshold criterion of statistical significance for strong relationships; and none of these strong correlations was large enough to conclude that spending alone exerts a pervasive influence on the nature of public services. The division between correlation directions was 20% positive and 18% negative. Sharkansky (1967) concluded that state spending is more often unrelated than it is related to outcomes. He also concluded that state spending seems counterproductive almost as often as it is productive.

Why would states that spend more receive less in benefits than states that spend less? Why do states that spend more per resident on police, criminal courts and incarceration experience higher crime rates? One possible explanation is that greater expenditures are always wasteful and ineffective. Standing in opposition to this explanation is the pattern that spending more per resident on police results in more police per resident. Are we to conclude that police cause crime? Can it be the case that fire fighters cause fire damage and public health employees cause infectious diseases? Rather than conclude that government efforts exacerbate problems, we can entertain the idea that government efforts are sometimes reacting to problems.

Reactive relationships between spending and results were as frequent as proactive relationships in Sharkansky’s study (Sharkansky, 1967). Moreover, both proactive and reactive relationships were found in each of the policy areas he studied: education, highways, public welfare, health and hospitals, natural resources. Sharkansky’s findings were even more complex: he reported both positive and negative correlations between many individual measures of results and different measures of spending.

The essential point is that the potential for government action to cause better results exists only in policy areas where governments can be proactive. As we contemplate the temporality of government efforts and problem levels we should be aware proactive and reactive relationships have different sequences and temporal modalities. For proactive relationships, government action precedes problem levels. For reactive relationships, government action follows problem levels.

Issue Two: Measuring money over time

Table 1 presents an example of an empirical relationship in the states that is consistently strong over the fifty year period of 1960 to 2010. There is an ongoing negative correlation between state welfare payments to recipients and state poverty rates. States that provide greater support to those in poverty experience lower poverty. If the relationship is causal, states might reduce poverty rates by increasing welfare payments. However, it could also be the case that states with lower poverty can afford to provide larger welfare payments. Causality could be in either direction. Further, the relationship could also be spurious, an artifact of one or more other state attributes.

OPEN IN VIEWER

Table 1.

Correlations between AFDC/TANF payments for families of four and state poverty rate

1960	−0.81		1988	−0.63		2000	−0.45
1970	−0.78		1989	−0.69		2001	−0.45
1975	−0.70		1990	−0.59		2002	−0.49
1979	−0.71		1991	−0.56		2003	−0.43
1980	−0.70		1992	−0.57		2004	−0.49
1981	−0.73		1993	−0.58		2005	−0.55
1982	−0.69		1994	−0.49		2006	−0.53
1983	−0.60		1995	−0.54		2007	−0.55
1984	−0.57		1996	−0.39		2008	−0.59
1985	−0.58		1997	−0.41		2009	−0.55
1986	−0.61		1998	−0.36		2010	−0.64
1987	−0.62		1999	−0.39

From 1960 to 1996 welfare payments were made through Aid to Families with Dependent Children (AFDC), a program established by the Social Security Act of 1935 to provide cash payments for needy children. ⁵ It was an entitlement program of federal funds supplemented, at each state’s discretion, with state funds. States defined ‘need’, set their own benefit levels, established (within federal limitations) income and resource limits and administered the program or supervised its administration. States were entitled to unlimited federal funds for reimbursement of benefit payments and were required to provide aid to all persons who were in classes eligible under federal law and whose income and resources were within state-set limits.

Under the welfare reform legislation of 1996, Temporary Aid to Needy Families (TANF) replaced AFDC, the Job Opportunities and Basic Skills Training (JOBS) program, and the Emergency Assistance (EA) program. The law ended federal entitlement to assistance and instead created TANF as a block grant that provides funds each year. The goals were to turn welfare into a program of temporary assistance and to assign responsibility for fashioning and implementing programs to the states. ⁶

The correlations between welfare payments and poverty rates in Table 1 show the impact of change from AFDC to TANF. Coefficients were very high from 1960 through the early 1980s, in the range of −0.70 to −0.80. Coefficient values then declined gradually until the reform legislation was passed in 1996. They were much smaller, −0.36 to −0.41, in the earliest years of TANF, 1996 to 2003 – perhaps years when states adjusted their policies. Thereafter, correlations increased again to levels seen from 1983 to 1995, −0.43 TO −0.64.

Motion Chart 1 animates the data underlying these correlations for the years 1960 to 2010. Circles representing states are the same size and colored to denote Southern (eleven former members of the Confederate States of America) or Non-Southern states. Welfare payments are measured in contemporary dollars to reflect information available to state officials at the time they made decisions. The Motion Chart 1 link is:

https://docs.google.com/file/d/0B3ffhHL7gNaqYlVNWWR4Tkl4bFU/edit?usp=sharing

Motion Chart 1 presents a general picture of welfare payments increasing and poverty rates declining over the 40-year period. Poverty rates for all states seem to fluctuate both up and down by approximately 3% every two years. Part of this back and forth movement could be interpreted as measurement error. There is a steady decrease in poverty from 1960 to 1974 and this suggests that so-called Great Society programs had the intended impact on poverty. Poverty rates decline for all states and then remain stable, with much smaller upward and downward movement.

States typically increase welfare payments and then hold the new level for several years. This makes the point visually that states do not reconsider welfare policy, or any policy, every year or every legislative session. The range of payments is consistent and smaller from 1960 to 1974; thereafter, while poverty is in a narrow range, the dispersion of welfare payment size becomes very large. It is noteworthy that most of the increase in the range of state welfare payments comes before the shift from AFDC to TANF. It seems the low correlation coefficients from 1996 to 2003 result more from a very low range of state poverty rates than from changing state benefit levels. Yet Motion Chart 1 shows that few states altered their payments to recipients after TANF changes were fully implemented, approximately 2001. The animation makes one think that the years of AFDC and TANF might best be thought of as distinct policy times even though the changeover does not seem a valid explanation for the historic low point of welfare-poverty correlations.

Motion Chart 2 visualizes welfare payments and poverty rates with payments in constant 2009 dollars instead of contemporary dollars. For each year, constant dollars are a simple multiple of contemporary dollars that reflects changing purchasing power of money. The correlations of Table 1 apply whether welfare payments are measured in contemporary or constant dollars. Although identical correlation coefficients apply to the animations in Motion Charts 1 and 2, the animations suggest quite dissimilar empirical phenomena. The Motion Chart 2 link is:

https://docs.google.com/file/d/0B3ffhHL7gNaqWlVGM0xuMkJ3SVU/edit?usp=sharing

Motion Chart 2 shows the same pattern of change in poverty rates as Motion Chart 1 but reveals a much different pattern on benefit levels. When benefits are measured in constant dollars, the major movement is backward, not forward. In terms of purchasing power, welfare benefits have declined. The approximate median monthly payment for a family of three has gone from $874 per month in 1960 to $600 per month in 2010. Similarly, the dispersion of benefit levels has decreased over the 50-year period, not increased.

States share a common pattern. Each increases its benefit level, then benefits immediately fall back to the previous level and then even lower. Periods of high inflation rates result in a rapid pace of backward movement and a narrower range of benefits that affect the states at the same time.

The change from AFDC to TANF in Motion Chart 2 is different from that in Motion Chart 1. The change to TANF came after a sustained period in which the purchasing power of benefits had declined in almost all states. States did not all increase benefits levels simultaneously and the impact of increases was relatively small. After the changeover to TANF was fully implemented in the states, around 2001, the purchasing power of welfare benefits consistently declined through the end of the time series. Motion Chart 2 does not seem to identify a critical juncture between distinct policy temporalities for welfare benefits. Instead, it shows an ongoing pattern of states moving forward and back, increasing the purchasing power of benefits then losing the gains to inflation.

The contrasting patterns of Motion Charts 1 and 2 illustrate the signal importance of measurement decisions. Welfare benefits eligibility and support levels are decided by state legislative and executive branch officials in terms of dollars and eligible individuals and families. Contemporary dollars reflect the information available when decisions are made. Motion Chart 1 can be viewed as a legitimate analysis.

Motion Chart 2 suggests that the analysis of Motion Chart 1 is unacceptably incomplete because Chart 1 ignores the importance of the changing value of money. It would be convenient if there were a simple rule of thumb specifying what length of time requires inflation adjustments to contemporary dollar measures. Unfortunately, the key element is change in the level of inflation. Large rates of inflation must be taken into account – and it does not matter whether large change occurs over decades or over a few years. Given the unpredictable pattern of inflation over time and space, the best advice is that adjustment for inflation is always good practice.

Motion Charts 1 and 2 both show some phenomena in common quite clearly. First, Alaska has become an extreme outlier in benefits provided. This is a result of the combination of a large influx of revenue from extraction of oil and continuing low population. Alaska becomes the leader in per capita and per recipient spending measures for almost all public policies and Alaska even has a negative income tax for permanent residents.

Second, the Southern states are always a distinct group. They consistently offer lower welfare benefits and experience higher poverty levels than the other states. Throughout the history of the USA, the Southern states have always funded education and welfare programs – and government services in general – at much lower levels than the other states. These differences reflect both a different view of the proper role of government and a seeming tolerance for higher poverty levels and lower educational attainment levels than exists elsewhere in the United States.

Issue Three: Expectations and norms changing over time

There is widespread agreement in the United States that educational attainment is an important goal. Moreover, there is broad agreement, both inside and outside the scholarly community, that educational attainment is linked to economic success, both for individuals and for communities (Day and Newburger, 2002; Baum et al., 2010). Educational attainment is also linked to better health and longer life expectancy (Braveman et al., 2010). It seems that there are no advocates of the position that lower educational attainment is better.

There are currently two threshold levels of educational attainment: completion of high school and completion of a bachelor’s degree. In the United States, higher educational attainment has become the key educational threshold – that is, the standard that current students are encouraged to meet for greater individual and community success. Three generations ago secondary school completion was that standard. Some empirical studies linking success and educational attainment use one measure of attainment or the other: some use both measures. Typically, higher educational attainment is more beneficial than secondary educational attainment, but not always. Analysis of the linkages between educational attainment and desired outcomes over time can help us identify which relationships have endured and which have changed.

The correlations between the two measures of educational attainment and state per capita income in Table 2 show the relationship is positive and strong across the states dating back to 1950. The pattern suggests that per capita income has become less strongly related to high school completion and more strongly related to completing a baccalaureate degree. This aggregate finding agrees with conventional wisdom for individuals and differential earnings cited above. Completion of a college degree has surpassed completion of high school as the key prerequisite for achieving higher levels of income. For individuals, the direction of causality is positive: education causes income. For states, the relationship could be education causes income, or income causes education, or both.

OPEN IN VIEWER

Table 2.

Correlations between state educational attainment of population aged 25 and older and per capita income

	High school	College		High school	College
1950	0.70	0.52	1999	0.34	0.69
1960	0.66	0.48	2000	0.30	0.75
1970	0.61	0.65	2001	0.33	0.78
1980	0.63	0.64	2002	0.38	0.77
1989	0.42	0.66	2003	0.39	0.77
1990	0.44	0.74	2004	0.36	0.75
1991	0.46	0.66	2005	0.37	0.75
1993	0.40	0.68	2006	0.30	0.73
1994	0.46	0.63	2007	0.37	0.75
1995	0.41	0.70	2008	0.36	0.75
1996	0.41	0.75	2009	0.38	0.78
1997	0.33	0.73	2010	0.35	0.79
1998	0.37	0.71

We also know that educational attainment in the states is empirically related to a number of other beneficial outcomes, including lower poverty rates (Table 3) and lower rates of infant mortality (Table 4). The negative correlations between educational attainment and poverty are larger and more consistent than the correlations between welfare and poverty in Table 1. The link between welfare and poverty is still important because, among other reasons, welfare levels can be changed when state officials choose to do so. Changing education levels would require more time.

OPEN IN VIEWER

Table 3.

Correlations between state educational attainment of population aged 25 and older and poverty rate

	High school	College		High school	College
1970	−0.72	−0.40	2000	−0.63	−0.53
1980	−0.68	−0.49	2001	−0.63	−0.57
1989	−0.66	−0.60	2002	−0.76	−0.58
1990	−0.69	−0.36	2003	−0.80	−0.60
1991	−0.66	−0.58	2004	−0.74	−0.60
1993	−0.74	−0.57	2005	−0.74	−0.62
1994	−0.69	−0.46	2006	−0.73	−0.57
1995	−0.66	−0.49	2007	−0.78	−0.58
1996	−0.64	−0.46	2008	−0.79	−0.61
1997	−0.63	−0.47	2009	−0.77	−0.57
1998	−0.65	−0.47	2010	−0.82	−0.65
1999	−0.63	−0.46

OPEN IN VIEWER

Table 4.

Correlations between state educational attainment of population aged 25 and older and infant mortality rate

	High school	College		High school	College
1950	−0.45	0.08	1997	−0.54	−0.42
1960	−0.34	0.07	1998	−0.47	−0.47
1970	−0.54	−0.21	1999	−0.51	−0.39
1980	−0.67	−0.26	2000	−0.51	−0.54
1989	−0.56	−0.43	2001	−0.37	−0.34
1990	−0.57	−0.41	2002	−0.50	−0.44
1991	−0.63	−0.52	2003	−0.45	−0.44
1993	−0.64	−0.50	2004	−0.46	−0.50
1994	−0.52	−0.36	2005	−0.53	−0.58
1995	−0.60	−0.50	2006	−0.52	−0.58
1996	−0.43	−0.39	2007	−0.56	−0.60

The link between higher educational attainment and income has surpassed the link between high school attainment and income. However, the correlations between poverty levels and high school attainment are consistently stronger. For infant mortality, the correlations with high school educational attainment are stronger through 2002. Thereafter, the two measures of educational attainment have indistinguishable correlations. Empirical links between educational attainment and income, poverty and infant mortality demonstrate all possible relationship patterns exist: higher educational attainment stronger, high school attainment stronger and both measures equally strong. For now, both measures of educational attainment are important positive correlates of outcomes preferred by everyone. ⁷

Table 5 presents the cross-sectional correlations between the two measures of educational attainment from 1950 to 2010. High school completion is measured as the rate of residents who have graduated from high school or the equivalent. College completion is measured as the rate of residents who have baccalaureate degrees – typically four years of college or university study – or a post-baccalaureate degree. For both measures of educational attainment the rate denominator is the number of state residents aged 25 and older.

OPEN IN VIEWER

Table 5.

Correlations between state population aged 25 and older and high school completion and baccalaureate degrees

1950	0.51	1995	0.59	2004	0.45
1960	0.50	1996	0.44	2005	0.49
1970	0.56	1997	0.48	2006	0.45
1980	0.60	1998	0.44	2007	0.53
1989	0.62	1999	0.48	2008	0.48
1990	0.60	2000	0.53	2009	0.47
1991	0.71	2001	0.44	2010	0.49
1993	0.59	2002	0.37
1994	0.60	2003	0.45

While the correlations between high school and college educational attainment are not the strongest, they are among the most consistent of any presented so far. With few exceptions, they are at or near 0.50 for the entire 60-year period 1950 to 2010. The consistent coefficients suggest the same relationships across the states are sustained for a long time. What can animated visualizations of these data add to our understanding? Will there be a single ongoing relationship confirmed or will there be changes and complexity? Motion Chart 3 shows that both simple and complex patterns are present.^8,9 The link for Motion Chart 3 is:

https://docs.google.com/file/d/0B3ffhHL7gNaqY1AxUms2NDd4Nmc/edit?usp=sharing

The first thing that is obvious from viewing Motion Chart 3 is all states exhibit impressive improvement on both measures from 1950 to 2010. Nationwide, high school completion increased from 34% to 87% and college completion increased from 6% to 30%.

A second clear pattern is that the Southern states are consistently grouped together with lower educational attainment than other states, on both measures, from 1950 to 1990. From 1990 to 2010, Southern states are more integrated with the other states but still underperforming as a group, although Virginia is an exception – it has become one of the states whose residents are the most highly educated. Georgia, North Carolina and Texas have separated themselves from other Southern states on higher educational attainment but still fall below the majority of states on both measures. Wyoming is a non-Southern and non-border state that has been consistently at or close to the bottom on both measures. Nevada, Indiana and New Mexico are non-Southern and non-border states that joined the Southern group at the bottom on both measures.

A third pattern evident in Motion Chart 3 is that the states move together as a group with little change in their relative positions on either measure of educational attainment for approximately 1950 to 1980. Having data for this early period only every ten years accounts for the smooth animated movement for each state. Having only decennial data does not account for the states maintaining their positions relative to each other. States exhibit important changes in their relative positions on both measures starting approximately in 1980. Motion Chart 4 animates the two measures of educational attainment for 1980 to 2010 to provide a more detailed picture of these years. The Motion Chart 4 link is:

https://docs.google.com/file/d/0B3ffhHL7gNaqVlZRa3c5OE5IQTQ/edit?usp=sharing

The tempo at which states change their positions relative to each other seems to become faster through the 30-year period. The short-term changes for some states have been dramatic. Texas is a case in point. Texas state officials can and do accurately boast that the high school completion rate increased from 63% in 1980 to 80% in 2009 and college completion increased from 19% to 26%. Motion Chart 3 shows that, in the same 30 years, twelve states surpassed Texas with regard to percentage of the population that completed high school. Texas went from ranking 38th to ranking 50th. The ranking of Texas with regard to college completions also declined, from 20th to 29th.

California exhibits a similar pattern on high school completion. Its rate went from 74% in 1980 to 81% in 2009, but its ranking dropped from 11th to 48th. California’s college completion rate increased from 21% to 30%, but its ranking dropped from 7th in 1980 to 14th in 2009. In Motion Chart 2, we can see that in 1950 California ranked second in high school completion and first in college completion.

The correlations between the two measures of educational attainment are nearly identical from 1996 to 2010; they are all within measurement error of each other. However, the states, while all moving ahead, are changing their relative positions as never before. It could be the case that the rankings are increasingly being affected by migration into and out of the states. Texas and California have both been surpassed by the other states in high school completion rates at least partly because each has retained, increased by birth and imported a large number of residents who have not completed high school. At the same time, these two states have become less enthusiastic about public education and have provided less in state support for public education. In both California and Texas the burdens of financing education have been shifted to local school districts that are constrained in their ability to raise tax revenue.

State educational attainment continues to be affected by state efforts to graduate their own young residents, but is also affected by population shifts. States actively compete with each other for businesses that employ highly educated people. We can use Motion Chart 4 to identify states that have been more successful in attracting college graduates. In 1980 there are approximately equal numbers of states, well separated from each other, above and below where a regression line would be. The states above that line have higher populations completing college than would be predicted from their populations completing high school. In 2009, there are ten states with much greater higher education attainment levels than would be predicted from their high school attainment levels: Massachusetts, Rhode Island, New York, New Jersey, Connecticut, Maryland, Virginia, California, Texas and Colorado. All are home to headquarters of large corporations and ‘knowledge industry employers’. They are prominently mentioned as states that actively seek and are attractive to highly educated people. Other states described similarly (for example, Illinois and Washington) are on the leading edge of the other 40 states.

Issue Four: Properly interrelating variables

Even if one remains skeptical that empirical relationships such as those shown in Tables 2, 3 and 4 are causal, the ongoing relationship between educational attainment and positive results raises other questions. Can state governments be proactive in achieving greater educational attainment among their residents? If so, what might be effective? Sharkansky’s (1967) question is germane: do states that spend more on education get more? His analysis of education spending and results in 1962 found significant positive correlations between primary and secondary education spending per capita and both measures of educational attainment. The correlation between spending and high school completion was 0.81. The correlation between spending and college completion was 0.66. Although spending per child would have been a better measure than spending per resident, Sharkansky’s findings are instructive. States that spend more might have greater educational attainment. Table 6 correlates educational attainment with state spending per child in public schools for the years 1960–2010.

OPEN IN VIEWER

Table 6.

Correlations between state expenditure per child and average daily attendance and educational attainment of population aged 25 and older

	High school	College		High school	College
1970	0.45	0.42	1999	0.23	0.42
1980	0.50	0.51	2000	0.21	0.46
1989	0.38	0.51	2001	0.20	0.48
1990	0.37	0.38	2002	0.25	0.48
1991	0.37	0.50	2003	0.24	0.46
1993	0.31	0.44	2004	0.25	0.44
1994	0.34	0.46	2005	0.25	0.47
1995	0.34	0.56	2006	0.25	0.54
1996	0.33	0.49	2007	0.32	0.52
1997	0.27	0.48	2008	0.30	0.53
1998	0.28	0.47	2009	0.38	0.52

The consistently strong correlations in Table 6 demonstrate educational spending and educational attainment are empirically related. But is it a causal relationship? If it is, the direction of causality must be that educational attainment causes educational spending, not vice versa. Educational attainment can cause education spending if, for example, more highly educated residents demand more spending in the public schools, both for their children and others’ children.

It is not possible for expenditures on education to cause rates of educational attainment in the same year. Contemporary spending on primary and secondary education promotes educational achievement for current students, ages 5–19. Educational achievement measures are for state population age 25 and older. The full impact of contemporary spending will take place when student beneficiaries of spending are all in the cohort used to measure educational achievement. Thus, a proper empirical analysis of educational spending on educational achievement must measure achievement 20 years after spending. Table 7 presents such an analysis.

OPEN IN VIEWER

Table 7.

Correlations between state spending per child and average daily attendance twenty years earlier and educational attainment of population aged 25 and older

	High school	College		High school	College
1970	0.70	0.47	1999	0.48	0.38
1980	0.66	0.48	2000	0.38	0.46
1989	0.51	0.49	2001	0.39	0.43
1990	0.47	0.44	2002	0.45	0.41
1991	0.49	0.48	2003	0.38	0.41
1993	0.39	0.52	2004	0.38	0.42
1994	0.42	0.52	2005	0.42	0.42
1995	0.38	0.42	2006	0.40	0.51
1996	0.50	0.58	2007	0.43	0.58
1997	0.55	0.48	2008	0.40	0.61
1998	0.49	0.42	2009	0.38	0.59

Lagged correlations between educational spending and high school educational attainment are consistently higher than contemporary correlations. On average, the correlation is 0.16 larger. Clearly, educational attainment is strongly and consistently linked to spending when the cohort of school children joins the cohort of adults whose educational attainment is measured. The correlations between lagged primary and secondary spending and college attainment are also usually higher, but the average difference is only 0.03. However, lagged spending correlations with higher education attainment greatly exceed those for high school attainment starting in 2007. Motion Chart 5 animates the relationship between lagged spending in constant 2009 dollars and completion of high school from 1970 to 2009; Motion Chart 6 animates the same relationship from 1990 to 2009. The links for Motion Charts 5 and 6 are:

Motion Chart 5

We have already seen that the states move together with little movement relative to each other on high school educational attainment from 1970 to 1980. Motion Chart 6 shows the same pattern. From time to time, individual states move ahead of others on spending, but then return to the previous positions. This again results from states increasing spending on education in some legislative sessions then not reconsidering until purchasing power has greatly declined.

Motion Charts 5 and 6 enable us to see that the states’ range of dispersion on high school completion became much smaller. In 1970, states with the highest and lowest high school attainment were separated by 22%: in 2009, the range was 12%. At the same time, dispersion on spending per child became much greater. In 1970, the range was $2,500; in 2009, it was $9,000. Alaska is not the only outlier on spending; New York, New Jersey and Connecticut spend nearly as much per child. These five states spend $12,000 or more per child. The other states spend between $4,500 and $10,500 per child. These are remarkable differences – particularly when we remind ourselves that all spending measures are normed to purchasing power in 2009.

As we have seen on the other Motion Charts, Southern states are grouped together at the bottom of both measures. They are joined with regard to extremely low spending per child by West Virginia, Kentucky, Oklahoma, Utah, Idaho, North Dakota, South Dakota, New Mexico and Arizona. These states have a different view of appropriate support for primary and secondary education. Of this group, only the Dakotas and Utah are above average in high school completion.

There are very strong and consistent correlations between what states spend on primary and secondary education and what states get in educational attainment when the beneficiaries of spending join the cohort measured for educational attainment. The link between what might be cause and what might be effect is separated by 20 years. This gives rise to at least two time issues. First, spending on primary and secondary education is long-term investment in human capital. Benefits come, but they do not come in the short term. Second, the time period for education return-on-investment and the political time horizons of elected government officials are severely disconnected. State and local officials cannot receive credit personally for good spending decisions or be held accountable for bad spending decisions.

Businesses are assessed quarterly and annually. A corresponding assessment period for state and local governments would be election cycles of two and four years. There is currently a very popular view that American state and local government should be run the way businesses are run. One of the implications of that view is that less emphasis should be placed on spending for long-term benefits, such as investment in replacing infrastructure and investment in future human capital. The Motion Charts on current and past spending and educational attainment identify several states that are pursuing a strategy of spending significantly less than other states on educating children in the public schools. Spending less need not result in lower educational attainment if these states are successful in luring people educated in other states and outside the United States to become residents.

The challenges of political time and policy temporality

Empirical analyses study temporal events. There are more timescape issues germane to studies involving multiple time frames than a single time frame. This paper has highlighted some noteworthy concerns:

Government policy may be proactive in affecting problem levels or reactive to problem levels. Sequencing and temporality modalities are obviously different for the two relationships. The directions of parameter estimates alone are not always sufficient to distinguish the two.

This paper has also demonstrated how animated visualization tools can usefully complement summary measures of association and other statistics. They provide considerable descriptive information in a simple intuitive format. Motion Charts convey all the intelligence inherent in single-time scatterplots and more. They also show change over time – knowledge that is fundamental to exploring phenomena such as temporality, timing, time frames and periodization, patterns, tempo, duration, sequence and critical junctures.

Motion Charts also communicate change over time in variable levels and dispersions. One can easily recognize outliers, but also change leaders, change followers and change resisters. Motion Charts clarify movement in time and space of both groups and the units that comprise groups. They provide considerable context to each item of information. They provide the means to confirm the claims of politicians in Texas and California about increases in proportions of residents who have completed high school. They also make it unavoidably clear that these states have been surpassed in high school educational attainment by almost all other states.

This paper has shown how animated visualizations of data can enhance understanding of the American states. We already knew the value of welfare payments measured in constant dollars has been declining, not increasing. Motion Charts suggest the timing and tempo of change may be influenced more by inflation rates than by state policy initiatives. The change from the AFDC program to the TANF program does not seem to be a critical juncture. States temporarily increased benefit levels, but did so at different points in time over a number of years.

We already knew educational attainment is positively linked to preferred outcomes for income, poverty, infant mortality and many other state characteristics not documented here. Motion Charts are not necessary to appreciate that the effect of spending on educational attainment of state residents requires decades to be noticed. Motion Charts do let us see the complexity of change in state educational attainment levels. While overall states are improving on both measures, the increased tempo of change starting in 1980 makes it clear that other factors have become important. One such factor may be interstate population relocation. Success in attracting residents educated elsewhere could be a viable explanation for why some states can fall behind others in graduating their own children from high school but maintain their rankings on higher education attainment.

Motion Charts visualize all individual cases simultaneously. They can be used as tools to identify units for intensive case-centered analyses – either because of their great change over time or their resistance to change. Motion Charts give valuable information about individual cases and also provide important intelligence about context – how individual cases relate to the bigger picture. Motion Charts could be useful as tools to assess the suitability of data for cross-sectional time-series analyses as suggested by Kittel (1999). Motion Charts might be a way to determine whether or not to use a constant-coefficients approach.

A most striking pattern common to all the animated data visualizations in this paper is the unmistakable regional distinctiveness between Southern and Non-Southern states. Civil war, economic depression, world wars and other key historical events have been recognized as critical junctures that created new eras for the American polity. However, the singularity of Southern state behavior remains. In an important sense, the differences across the American states mirror the same patterns present at the creation of American federalism. American society continues its struggle to distinguish state differences that are desirable, beneficial, benign, unavoidable and tolerable from those that are not.