Dissipation factor seems to be the figure of merit being bandied about for crossover caps. Although DF has some importance I'd like to set the record straight, while hopefully keeping the explanation simple.

Dissipation factor is defined as,

    DF = ωRC × 100%, where ω = 2πf

DF is the ratio of resistance to capacitive reactance and therefore represents loss. Our hearing is able to discern relative sound levels only within about 0.3 dB. This corresponds to a change of 3.5%.

A metalized polyester cap with a DF of 0.25 at 1 kHz will have a DF of 2.5 at 10 kHz, a difference of 2.25% between 1kHz and 10 kHz. DF is a non-issue when used in series with the high frequency driver (the typical high pass circuit) since the series resistance of the cap adds to the series resistance of the driver and is effectively swamped by it.

A 5μF capacitor with a DF of 0.25 at 1 kHz has an equivalent series resistance (ESL) of just 0.08Ω, or 1% of the resistance of an 8Ω driver. Since the ESL of the cap is constant it has no more of an effect than a resistor in series with the driver.

When used as a high pass shunt element (band or low pass circuits), the ESL of a cap will limit the maximum attenuation achievable at higher frequencies. As an example, with a high band cut-off of 2 kHz, the typical DF error (based on a cap of 5μF, DF of 0.25, an 8Ω driver and 12 dB per octave) is in the range of 5% at 20 kHz, or 0.42 dB.

Although this means the mid or bass driver is stronger in it's output by 0.42 dB at 20 kHz, it doesn't mean this difference will be audible. The high driver signal is already about 40 dB stronger than the mid or bass driver at this point.

More important to perceived audio quality is equivalent series inductance (ESL). Introduce ESL and the phase relationships will switch directions (whether used as series pass or shunt) as the signal approaches the higher frequencies. In practical terms phase changes affect the arrival time of sound to our ears.

Since our hearing is far more sensitive to timing relationships than sound levels, these phase errors will be far more noticeable. Physcoacoustic studies have established about a 10 dB preference being given to similar sounds which arrive at our ears 1st. This means that if the 1st sound is 10 dB fainter than the succeeding one the fainter sound is just perceptible in location, otherwise it blends in with the stronger sound.

High ESL is the main reason you should replace, or at least bypass, non-polar electrolytics in your newer crossovers. The electrolytic caps in older crossovers may have already dried out and shifted their capacitance, so they should always be replaced. Basically any low inductance wound metalized polyester cap of the same capacitance and voltage will do, but you should get caps which are fully sealed so they won't be affected by high humidity conditions. Many of the older tape-wound axial caps are not fully sealed.

Parallel two caps where the leads enter the PC board to lower inductance even further. For instance, if you need 4.4μF then parallel connect two 2.2μF caps.